On the last day of April, 2021, a new shareholder appeared in the corporate records of Beijing ByteDance Technology, the subsidiary that holds the licenses allowing ByteDance to push video and news onto a few hundred million phones across China.

A week earlier, it had not existed. It was called WangTou Zhongwen, a name that rendered in English as something very forgettable: Net Investment Chinese Media. No one could locate its offices or name its staff. The stake it bought came to one percent, and the price, when the reporting finally pried it loose, was about 2 million yuan (roughly 290,000 dollars). The company, now part-owned, was worth a quarter of a trillion. The 2 million yuan was loose change. The one percent came with a seat (one of three on the board) plus a veto over the only two questions that decide what a company like ByteDance is: what the platform shows the people who open it, and where it gets the money to keep going. The seat went further still; scholars who later took the structure apart found it resting a hand even on the choice of the editor-in-chief.

The arrangement has a name in Chinese: 特殊管理股, the special management share. By the time it reached ByteDance it had been circulating through the country’s media companies for half a decade, always in the same plain costume. A state-backed fund takes a sliver of equity, around one percent. With it comes a director, and a veto over the short list of decisions the Party minds most. Weibo had disclosed a nearly identical setup to American securities regulators a year before. Kuaishou had one. So did corners of Alibaba and Tencent. In none of these cases did the state buy the company. It took a chair and a veto and left the rest of the shares where it found them, in private hands, on a private balance sheet, held by people who were now, in the ways that counted, no longer in charge of them.

For years, the special management share read as a foreign specimen, the kind of quiet instrument that surfaces where the Party sits inside the firm and never where it does not. People who track these things took it exactly that way. Pablo Chavez, who has run public policy for Google, Microsoft, and LinkedIn, read about the ByteDance stake and set the reflex down almost word for word: it looked, he wrote, like something Beijing would do and Washington never would.

He turned out to be wrong. Over a single year the United States reached for the same kind of control, deliberately, and in the open. It did so as a direct answer to China: industrial tools Washington had treated as foreign, even un-American, for the better part of a century, picked back up because it had concluded that a country which keeps both hands off the wheel cannot compete with one that does not. The instrument did not migrate home on its own. A government went and got it.

The story of why Beijing wanted so quiet an instrument runs through a man who was the opposite of quiet, on a stage on the Shanghai waterfront, in the autumn of 2020.

On the evening of the 24th of October, Jack Ma walked out in front of the Bund Summit, a finance conference whose audience was thick with the central bankers and regulators who run the Chinese economy. He was the most celebrated businessman in the country, a former English teacher who had built Alibaba into a colossus. In nine days he would take its payments arm, Ant Group, public in a dual listing in Shanghai and Hong Kong, an offering set to raise around 37 billion dollars and value the company north of 310 billion. It was the largest in the history of the world. The order book came in oversubscribed roughly 870 times. The deal was a triumph before he opened his mouth, and the moment asked only for grace. Instead he told the regulators sitting in front of him that they were doing it all wrong. China’s banks ran like pawn shops, he said. The country carried no systemic financial risk because it had no functioning financial system to speak of. Innovation was dying under small men in love with their own paperwork.

Nine days later there was no offering. On the second of November the authorities summoned Ma to what they delicately called a supervisory interview, a joint sitting with the central bank and three other regulators. The next afternoon, with the stock 36 hours from trading, the Shanghai exchange suspended the listing, citing a sudden and unspecified change in the regulatory weather. Hong Kong followed within hours. The Wall Street Journal later reported that the order to kill the deal had come from Xi Jinping himself. Ma had been the nearest thing China had to a folk hero of private enterprise. He more or less vanished. Over the next year the campaign against the platform companies erased well over a trillion dollars of their combined value. Didi, the ride-hailing company, had pushed ahead with a New York listing against the regulators’ plain wishes; it was hounded into delisting within months.

None of this taught Beijing to put the hammer down. The property developers and the tutoring companies would find that out within the year. China’s version of Kim Kardashian, later still. But the hammer, for everything it could flatten, could never do one thing: it could not sit inside a company and stay there. A crackdown lands and then lifts; it does not hold a standing veto over what a platform shows tomorrow morning, or where it raises its next round of money. For that the state needed an instrument it could leave in place, quiet and permanent, reaching only the decisions it cared about and none of the rest. It needed a scalpel to go with the hammer.

The crackdown ended the thing it aimed at. It also did something the planners had not put a price on. It taught private capital that no fortune in China was ever entirely private, that a regulator’s mood could erase a decade of work between a Friday and a Monday. Money does not forget a lesson like that. The campaign sparked what one account later called a yearslong chill across Chinese technology, and the founders who could leave began to look for the door. Keeping a hand on the companies that mattered, without staging another public execution, was the problem Beijing spent the next several years trying to solve (with mixed results).

But the scalpel was already lying on the table. In 2016 the broadcast regulator had called the heads of the big streaming services into a closed room and told them to receive a state stake of one percent or more, a director, and a right to review their content. The next year the Party’s central office and the State Council issued formal guidance piloting the special management share across online news, publishing, and video. By 2021 the cyberspace regulator’s investment fund was taking its quiet one-percent positions across all the platforms that mattered. At ByteDance, three state entities ended up holding the seat at once: the cyberspace fund, a media group run out of the Party’s propaganda department, and the Beijing city government’s investment arm. No headlines marked any of it. A shell company appeared on a filing. A director took a chair. The state let itself in near the head of the table, close enough to decide who edited the news.

The instrument was not even Beijing’s invention. The special management share was minted in Margaret Thatcher’s Britain, in the privatization sales of the 1980s, to solve an open-market problem: how to hand a state asset to private buyers and keep one hand on the wheel. The government sold the companies and kept, in the ones it cared about, a single special share worth a face value of one pound. It paid no dividend and carried almost no ownership. What it carried was a veto, over a foreign takeover, say, or the sale of a firm that built weapons. Britain still holds golden shares in BAE Systems and Rolls-Royce, the firms that make its warships and the reactors inside its submarines. The logic was plain from the first day. Sell the company; keep the veto. One pound was always enough, because the power in the share had never been tied to the money behind it.

Then Europe spent 20 years stress-testing it. The European Commission decided a government veto hidden inside a privatized company scared off investors from other member states, and it hauled one government after another to the European Court of Justice. The Court struck down the golden shares of France and Portugal, then Spain’s, then the German law that had let the state block a takeover of Volkswagen. One share survived the cull. It was Belgium’s, held over the companies that moved the country’s gas and oil, and it survived for one reason: its powers rested on “precise criteria which are known in advance,” written down, open to challenge before a judge. France’s and Portugal’s were a minister’s prerogative, exercised behind a closed door, and that was the thing the judges would not allow. The version of state control European law could live with was the one you could see.

By the time Washington reached for it, the tool had been tested to destruction on three continents, and the test had returned one clear result: the only state control a free society proved able to tolerate was the kind it could see. The United States was about to run the test again, live, at scale, and largely in public.

The settled wisdom held that it would never reach Washington at all. Almost every other country, rich or poor, keeps a national airline the state owns or stands behind (Singapore Airlines, Emirates), a flag in the sky and a guaranteed line to the rest of the world. The United States never built one, and never thought it should. Its self-image ran on the opposite premise: the government did not own the means of production, did not take seats on corporate boards, and, outside of wartime, did not hold a veto over what a private company sold or where it built a factory. Then, in a single year, it did all three, in the open.

The intent was on paper by the second month of the year. On the third of February, 2025, the President signed an executive order directing the Treasury and Commerce secretaries to draw up a plan for a United States sovereign wealth fund. The aim was to do for the federal balance sheet what Norway does with its oil, to “monetize the asset side of the U.S. balance sheet,” in the Treasury secretary’s phrase. Howard Lutnick, the commerce secretary, a blunt former Wall Street trader who would turn up again and again in the year’s deals, mused that the fund might one day take a profit-earning stake in vaccine makers. The President floated it as the place to park a future American stake in TikTok, the very company whose Chinese parent had taught (or reminded, rather) everyone what a special management share was. The fund has yet to appear. The appetite behind it has, though, over and over.

In June, the government cleared the Japanese conglomerate Nippon Steel to take over U.S. Steel, a deal the previous administration had blocked outright, on one condition: it would receive a golden share, the first the federal government had ever held. The device let an American icon pass to a foreign owner without anyone having to say so plainly. The buyer got the steelmaker. The government kept the veto. And what a veto! It covered closing plants, moving production abroad, reducing Nippon’s promised investment, shifting the headquarters out of Pittsburgh, and even changing the company’s name. It allowed the President to personally appoint one of the three independent directors and block the other two. Every line of that control was written down, in a deal anyone could read. When U.S. Steel moved to idle its works in Granite City, Illinois, and let 800 workers go, the Commerce Department reached for the share and reversed the call. “We have a golden share, which I control,” the President told reporters, pleased. The share has no expiration date. When this administration leaves, it passes intact to the Treasury and Commerce departments, to whoever holds those offices next.

The Intel deal arrived two months later.

Lip-Bu Tan was an unlikely man to find at the center of a national-security drama. He was born in 1959 in Muar, a port town in what was then the Federation of Malaya, into an ethnic Chinese family. He studied physics in Singapore, then crossed to the United States in the 1970s for a doctorate in nuclear engineering at MIT. The meltdown at Three Mile Island soured him on the field, by his own telling, and he dropped out and drifted to Silicon Valley. In 1987, he started a venture firm, Walden International, with 3 million dollars put up by family friends. Over the next 30-odd years he built it into a machine running more than 5 billion dollars and holding stakes in some 600 companies, more than 100 of them in China. One early bet, in 2001, was a seed check to a young foundry called Semiconductor Manufacturing International Corporation, now the largest chipmaker in China; he sat on its board for the better part of two decades. Another was Sina, which went on to create Weibo, which would in time sell its own special management share to the state. In March of 2025, after the board pushed out his predecessor, Tan took the top job at Intel. The company was an American monument in steep decline, its stock down by more than half over the prior year, 15,000 jobs already cut. “Intel is where good reputations go to die,” a veteran Valley executive and friend of Tan’s told the Irish Times, by way of congratulations.

At 4:39 in the morning, Pacific time, on the seventh of August, with Tan presumably asleep in California, the president posted that Intel’s chief executive was “highly CONFLICTED and must resign, immediately.” The trigger was a letter from Senator Tom Cotton, the Arkansas Republican, to Intel’s chairman. It flagged Tan’s tangle of investments in Chinese chip firms, some reportedly tied to the People’s Liberation Army, and noted that Tan had run the design-software company Cadence during a stretch when it sold technology to a Chinese military university, an episode Cadence had recently pleaded guilty to. The two men had never met. The heads of Nvidia and AMD and OpenAI and Google had all made their pilgrimages to the White House. The head of America’s oldest chipmaker had stayed in Santa Clara, and had not given to a presidential campaign in over two decades. He had simply lacked the antenna, one analyst said, to see it coming. Intel’s stock slid on the post. The company put out a stiff line about its commitment to American security.

Four days later the weather reversed. The President received Tan in the Oval Office, flanked by Lutnick and the Treasury secretary, and came out calling the man’s career an “amazing story.” Eleven days after that, on the 22nd of August, the government took a 9.9 percent stake in Intel, buying 433.3 million shares at 20.47 dollars apiece. The whole thing was paid for by converting chip-factory grants Intel had been promised but not yet handed. A warrant let the government buy another five percent if Intel ever lost control of its foundry business. All of it sat in a public filing, the exact share count and the exact price. Fifteen days. In that span, the man the president had demanded resign over his closeness to China became a partner in whose company the United States government was now the largest single shareholder. The accusation had become the acquisition. The president boasted that the shares had cost the taxpayer nothing, an arithmetic that worked only if the grants he had threatened to claw back were counted as money that never existed.

The Nvidia arrangement, that same August, skipped the equity and went straight for the till.

Jensen Huang, Nvidia’s founder, is a showman in a black leather jacket who had spent the better part of a year courting an administration he had been slow to embrace. He was the one major technology chief absent from the inauguration. The courtship began, by his own account, with a cold call from Lutnick, who introduced himself with a formality Huang later found funny enough to repeat on a podcast, and an invitation to come to Florida and sing for his supper. In the spring Huang paid a million dollars a head to dine at Mar-a-Lago, and spent the evening over gold-rimmed plates pressing the President not to choke off his sales to China, a market then worth tens of billions to one of the most valuable companies on earth. He does not take a setback gracefully. Later pressed on the China question, he nearly lost his composure, snapping that he was not someone who had woken up a loser.

What he got was a release on a leash. The administration let Nvidia and AMD resume selling a China-specific chip, the H20, that it had banned in the spring, in exchange for 15 percent of the revenue from those sales, paid to the Treasury, as the price of an export license. Export licenses do not normally carry a fee. The President, who had taken to calling the chip old and obsolete, opened at 20 percent and let Huang talk him down to 15. “I’m giving you a release,” he said. Deborah Elms, a trade-policy analyst at the Hinrich Foundation, named the obvious objection. “You either have a national security problem or you don’t,” she said. A 15 percent cut does not make the danger disappear. It gives the government a stake in the thing it had called dangerous. That is a different posture entirely.

A month earlier the Pentagon had taken a stake of roughly 15 percent in MP Materials, the country’s main rare-earth miner, and become its largest shareholder. The deal set a floor price of 110 dollars a kilogram for the two rare-earth oxides the company sells, close to double the going rate. When the market ran low, the government would pay the gap each quarter. When it ran high, the government would take 30 percent of the upside. It is, almost line for line, the guaranteed-price-and-state-stake structure China had spent two decades perfecting. Until 2022, MP had sold nearly all of its raw output to Shenghe, a Chinese rare-earth company that remains one of its largest shareholders. The American escape from dependence on China was financed by copying China’s method into a company whose biggest customer had been Chinese. This is the case hardest to argue with. A rare-earth supply chain is a chokepoint, and China spent two decades buying up the world’s. When a rival holds the chokepoint and has already shown it will squeeze, keeping both hands off your own scale just leaves you holding nothing. The market that was supposed to allocate the metal had been bent years before, by the other side.

By autumn the habit had reached the ground itself. In October the Energy Department took a five percent stake in Lithium Americas, plus another five percent of the Nevada mine it is digging, for softening the terms of a federal loan. A week later the Defense Department took a 10 percent stake in Trilogy Metals, which wants to pull copper and cobalt out of a remote corner of northwestern Alaska; in exchange the administration agreed to approve a 200-mile access road through wilderness that had been rejected, on environmental grounds, before. None of these was a rescue. That is what made them new. The government had reached for corporate equity before, in the wreckage of 2008, when the Treasury took preferred stock in the banks and the carmakers and sold it back the moment the patient could stand. This ran the other way. The government was buying into, as one researcher in the field put it, “an industry that has not yet launched,” taking positions in things not yet built.

I have argued before that power does not vanish when a new system starts to matter; it rebinds onto it. The state did not nationalize steel or silicon or lithium or rare earth. It left the ownership formally untouched and fastened a fresh hold onto each one anyway: a director here, a cut of the revenue there, a price floor and a guaranteed buyer somewhere else, a warrant that trips if a company stops behaving. The instrument that was supposed to belong to the other side described Washington’s year with uncomfortable precision. And one feature ran through all of it. Every stake had a price, every veto a filing, every trigger spelled out in advance. The government had let itself into a dozen boardrooms, and you could trace, deal by deal, exactly how it got into each one.

None of this was drift. Each move was a choice, and the choice had a target. Washington had watched a state-directed rival outbuild it in steel, in solar, and in the metals that go into both, and decided that a country which keeps both hands off the wheel cannot stay in a race against one that does not. The conclusion may even be right. The trouble is that the method is going badly for the country that wrote it.

China’s strongest decades were its least directed ones, the long stretch after 1978 when the state mostly stepped back and let private firms and foreign capital pull the better part of a billion people out of poverty. The turn back toward control has not paid the way the turn away from it did. Beijing has spent years trying to be strong in everything at once, subsidizing every strategic sector, leaning on its own firms, and squeezing the household spending that a normal economy runs on, in an attempt to force into being a kind of economy that has never existed: a manufacturing superpower that never has to sell very much to its own people. The bill for that arrives slowly, in weak demand and soft confidence and capital that would rather be somewhere else. The crackdown that handed Beijing its quiet instrument was one line on that bill. The country that wrote the method is the one now paying for it.

On the 18th of June, 2026, Senator Bernie Sanders introduced the American AI Sovereign Wealth Fund Act. It would levy a one-time 50 percent tax, payable in stock, on artificial-intelligence companies with more than 200 million dollars in sales. The shares would drop into a sovereign wealth fund. The votes that came with them would go to an independent commission empowered to block company decisions it judged harmful and push the ones it favored. The dividends would go to American households, on the model of the checks Alaska cuts its residents from its oil fund. To the populist left, this was the public taking back what the public’s data had built. The same month, from the far end of the spectrum, the president told reporters he would soon sit down with the AI companies about federal stakes, sketching an outcome in which “the American public essentially becomes a partner.” A democratic socialist and a Republican President, in the same month, reached for the identical instrument: voting shares and a seat on the board, used to block. Each was convinced he was rescuing the country from the other one’s idea.

The two boldest moves either government made all year landed on the same industry, six weeks apart, and neither one broke a law. Both reached into ordinary, legal, voluntary activity and simply stopped it. The people who watch this closely are still arguing about what they saw.

Start with Beijing, in the spring. In December of 2025, Meta had agreed to buy an AI startup called Manus for more than 2 billion dollars, the clean exit that founders across Chinese tech had been waiting for, proof that you could build a company in China and cash out to an American giant. Manus had taken great pains to be acquirable. Built in Wuhan in 2022, it had gone viral in March of 2025 on a demo of an AI agent that carried out long tasks on its own, and by the following summer it had moved its headquarters to Singapore, laid off most of its China staff, cleared its Chinese social accounts, and blocked mainland IP addresses—the full routine a Chinese company performs to read as a global one. On paper, it was Singaporean. But none of it mattered. On the 27th of April, 2026, China’s economic planning agency, through the office that runs its foreign-investment security review, ordered the deal unwound, the first foreign acquisition in artificial intelligence it had ever publicly blocked. The decision reportedly went up to the national-security commission that Xi Jinping chairs. Officials reviewing it described the sale, in the reporting that followed, as an attempt to hollow out the country’s technology base. The thing Beijing was protecting sat underneath the letterhead: the engineering, the training data, and the people—all built on the mainland, none of which it would let walk out the door under another flag.

Washington’s version came six weeks later, and it reached … somehow further.

Dario Amodei makes an unlikely insurgent. He is a physicist, soft-spoken and a little owlish, who led the teams behind GPT-2 and GPT-3 at OpenAI. In December of 2020 he walked out, with his sister Daniela and a handful of colleagues, over what they took to be the organization’s (you would not call it a “company” back then) cavalier attitude toward safety. The lab they built in response, Anthropic, made the Claude models. Amodei spent the next several years in a posture almost nobody else in the industry could hold for long. He wrote a 15,000-word essay imagining that powerful AI might cure most disease and double the human lifespan. In the same breath, and the same calm register, he warned that it might also slip its makers and end very badly. He believed, an admiring reviewer wrote, that the industry was building gods, and that humanity might not survive the building. On a Sunday news program in late 2025 he said that he was “deeply uncomfortable with these decisions being made by a few.”

He meant the company chiefs. He did not yet know how completely the government was about to prove that the “few” people in question were not the ones he had in mind.

Through the early months of 2026, the Defense Department negotiated with Anthropic over the terms on which the military could use Claude. The talks foundered on two points the company would not give up. Anthropic was, by its own account, willing to support nearly any lawful national-security use of its models, with two exceptions: It would not let Claude be turned on the mass surveillance of Americans, nor would it let it run fully autonomous weapons that chose and killed with no human in the loop. The Pentagon wanted access for “all lawful purposes,” and argued that a private contractor had no business telling the government how to use its tools in an emergency. One version the Pentagon floated, according to reporting on the failed talks, would have required access to Americans’ geolocation and financial records, pulled from commercial data brokers.

On the 27th of February, Pete Hegseth, the Secretary of War, a title his department had recently revived from the first half of the 20th century by executive order, announced on social media that he was designating Anthropic a supply-chain risk, a label ordinarily reserved for companies tied to foreign adversaries. The President ordered every federal agency to stop using the company’s technology at once. Anthropic, which said the two exceptions had not blocked a single government mission to that point, sued. In late March a federal judge in San Francisco, Rita Lin, granted an injunction, writing in a blistering opinion that nothing in the law supported the “Orwellian notion“ that an American company could be branded a saboteur of the United States for disagreeing with its government. In April a federal appeals court declined to block the Pentagon in Anthropic’s separate, narrower case, even as Lin’s injunction held. The fight ground on into the summer.

Among the reasons the government gave, in its own court filings, for treating Anthropic as a danger to the nation was this: the company might be able to disable or quietly alter its own model in the middle of a live military operation, if it decided its policies were being crossed. A private firm, in other words, might hold a kill switch over its own technology. The government found that intolerable.

Then, on the afternoon of Friday the 12th of June, at 5:21 Eastern time, the government reached in and threw the switch itself. An export-control directive, citing national-security authority and ostensibly worried about a method of jailbreaking the safeguards on one model, ordered Anthropic to cut off all access to its two most capable systems, Fable 5 and Mythos 5, for every foreign national on earth, including the company’s own foreign-national employees. Anthropic had no way (yet) to sort its users by passport, so it switched both models off for every customer in the world within hours. Weeks earlier the same systems had captivated Wall Street and Washington. By the company’s own account, the letter that turned them off “did not provide specific details“ of the concern behind it.

Set the two side by side. One government stopped a sale; the other switched off a product. One invoked a foreign-investment review; the other invoked export control. Both reached into activity that was legal, voluntary, and already underway, and stopped it cold on the authority of national security, and in neither case had the company on the receiving end broken a law. Meta said the Manus deal “complied fully with applicable law.” Anthropic had been running two products the government itself was using days earlier. No property was seized. No crime was charged. What happened was newer than either, and harder to name: a state deciding that a particular piece of artificial intelligence was too important to be left to the people who owned it, and moving it where the market never would have. Two governments, six weeks apart, at opposite ends of the earth, arrived at the same decision, and reached for it with the same hand.

The honest position starts by conceding the thing the libertarians will not. The state’s grip on the country’s most important companies is now something both parties want (the year just past says so plainly), and against a rival that builds this way, it may simply be the cost of staying in the race.

Where the grip stays narrow, it is defensible. But there is one mistake the United States cannot afford to copy, and it is the one China made. Top-down management has a failure mode that has nothing to do with which sectors a government picks. It frightens the people whose money it needs. Capital that no longer trusts the rules leaves, or hides, or waits; the state moves into the space it vacates; the vacancy frightens the next tranche of capital, and the thing feeds on itself. Confidence is slow to build and quick to break, and breaking it is the surest way to turn a selective intervention into a general retreat. China is several years into learning this. The United States is at the beginning.

So the form the grip takes decides everything. The equity stake, the thing the libertarian and the hawk both fix their eyes on, is the part that behaves itself. A named director, a disclosed price, a veto exercised somewhere a court can see it: that kind of control is selective by construction, because it has to be argued for in the open and can be fought there. It was the only kind Europe’s judges would let stand. The dangerous instrument is the other one—the one with no filing, no price, and no reason given. China’s review of the Manus deal ran largely in the dark, on grounds it never had to fully spell out. The American letter that switched off two models on a Friday gave no reason at all. Power that leaves no mark on a balance sheet is the more total kind, precisely because there is nothing on the record to point at, and nothing on the record is exactly what teaches capital to be afraid.

The position worth taking is the one almost no one is taking. The argument that the state should be kept out is lost; both parties walked away from it. What is still live is the form. When the state comes in, it should come in an expensive, visible way. Make it buy its share. Sit its director in a chair with a name on it. Put its veto on the record, where the price can be counted and the call fought in open court. Where the intervention is genuinely forced, as it is at a weaponized chokepoint like rare earths, the legible version is the right answer, and a defensible one. The instrument to refuse is the silent letter, because the silent letter is how a country talks itself, one unaccountable decision at a time, into the arbitrariness that empties a room of capital.

None of this is new, which is the unsettling part. Britain built the instrument, and Europe stress-tested it and kept only the version that stayed in the light. China picked it up, reached past it for the blunter tools lying beside it, and is now several years into paying for what those tools did to the confidence of the people it needs most. The United States has spent a single year reaching for the same things, in the same order, under the same logic of necessity, and it has already begun to prefer the quiet version: the letter over the share, the decision no one has to defend.

This has happened before. Twice this spring, in two capitals that agree on almost nothing, a government looked at a piece of the future, decided it was too important to leave in private hands, and took the wheel away from private machinations. The metals and the models are different. The instinct is identical, and so is the long record of what it costs a country when this kind of power stops being selective and stops being seen. The tool can be picked up. It has to be picked up slowly, narrowly, and in daylight, with both eyes on the one mistake that turns an instrument of competition into the thing that hollows a country out from the inside. The warning is not that the state takes a hand. It is what happens when the hand stops leaving fingerprints.

In a Beijing classroom in August 1966, a sixteen-year-old girl named Song Binbin watched her classmates beat the vice principal of the Girls’ Middle School Attached to Beijing Normal University to death with belt buckles, wooden boards, and clubs.

The vice principal, Bian Zhongyun, was fifty years old, the mother of four children, and had taught at the school for seventeen years. Her students had been her students that morning. By evening they had killed her in the school’s courtyard, and her body lay there for hours because nobody could be found who was willing to move it.

Calling those teenagers “psychopaths” is rhetorically satisfying and analytically lazy. The much harder and more useful reality is that they were responding to a coherent set of incentives. The class-status system that had guaranteed their university placements was eroding under reforms that rewarded competence over family background. Normal career paths had been suspended. Most consequentially, by late 1966, not participating in struggle sessions had become its own marker of suspicion.

The choice was between persecutor and persecuted, and the space between those positions had collapsed to nothing.

Two weeks after Bian Zhongyun’s death, Song Binbin stood on the rostrum at Tiananmen Square and pinned a Red Guard armband on Mao Zedong’s sleeve. The Chairman asked her name. Binbin meant gentle, refined. He suggested she change it to Yaowu, meaning be violent. She did, for a time.

The ideology was real. But the energy came from somewhere else.

A few weeks ago, considering the sort of awkward ideological interregnum American society currently finds itself in, I got caught up in the thought of trying to understand what happens when paradigms “die,” and the pattern is unsettlingly consistent across centuries, continents, and levels of violence. The Cultural Revolution is one example, and the most extreme one I will examine here, but it is not unique.

The same underlying structure appears in cases as different from each other as the collapse of imperial China, the violent dismantling of the samurai class in Meiji Japan, and the Soviet enforcement of pseudo-science against Mendelian genetics.

They share an architecture. A coalition senses its frame is slipping. It escalates enforcement against defectors. It expands the definition of heresy. It merges intellectual disagreement with moral disqualification. And in doing so, it accelerates the very collapse it is trying to prevent.

We are living inside one of these transitions right now. Almost nobody is describing it clearly. The work below is an attempt at a description.

A Paradigm Is a Coalition

In 1962, a philosopher of science named Thomas Kuhn published The Structure of Scientific Revolutions, which introduced an idea that has since escaped its original context and entered general use: the paradigm shift.

Kuhn’s argument was that science progresses through long periods of stable consensus (which he called “normal science”) punctuated by abrupt transitions when the reigning framework can no longer absorb its anomalies. The Ptolemaic model giving way to Copernicus. Newtonian mechanics giving way to Einstein. The old framework gets replaced wholesale, often over fierce resistance from practitioners who built their careers inside it.

Kuhn’s account of that resistance was primarily intellectual. Practitioners reject anomalies, add epicycles, and refuse to engage with rival explanations.

He underestimated his own theory.

The historical record shows the defense is almost always more aggressive than that, because a paradigm is a coalition first and a set of ideas second. The coalition’s members have built careers, institutions, reputations, and self-conceptions around the paradigm being authoritative.

When the paradigm is secure, the costs of membership are low and the rewards are high. As it weakens, the cost-benefit shifts, and every marginal member has increasing incentive to defect to whatever the successor framework will be. The coalition’s leadership notices the defections. It recognizes that defection past a threshold becomes irreversible. And it increases the cost of defecting through social, professional, and sometimes legal sanction. The escalation is entirely rational. A coalition trying to prevent its own collapse by making departure more expensive than loyalty.

What I have been trying to understand is the architecture of that escalation, because the historical cases share so much structure that they practically constitute their own minor field of study.

Three I will examine here in detail are the late Qing dynasty’s failed reforms between 1898 and 1912, the Meiji oligarchy’s violent dissolution of the samurai class culminating in the Satsuma Rebellion of 1877, and Lysenkoism in the Soviet Union from the 1940s through 1964.

Each shows the same architecture operating in radically different cultural and political contexts. Each ends in the collapse the defense was meant to prevent.

The Five Moves of a Dying Paradigm

Late-stage paradigm defense takes characteristic forms, and they are consistent enough across cases that you can practically use them as a diagnostic checklist.

The first is the expansion of the heresy category. As the paradigm weakens, more positions get coded as heretical, often positions that were previously tolerated within the paradigm itself. The boundary of acceptable disagreement narrows, and the narrowing is deliberate. A coalition compressing the space within which defection can happen by relabeling moderate positions as extreme. The compression tracks the perceived threat and bears little relationship to the actual content of the disagreement.

The second is the merger of intellectual and moral categories. When the paradigm was secure, disagreement was coded as wrong but not necessarily evil. As it weakens, disagreement gets coded as morally disqualifying. Heresy becomes an unforgivable sin rather than the annoyance it might have been considered before. The intellectual content of the disputes does not change nearly as much as the moral coding around them. The moral coding is doing the enforcement work that confidence used to do.

The third is the escalation of professional consequence. Early-stage paradigms tolerate disagreement because they can absorb it. Late-stage paradigms cannot tolerate disagreement because each visible dissident demonstrates to potential defectors that defection is survivable. The result is a sharp increase in the cost of dissent. Lost positions, lost funding, lost standing, sometimes lost freedom or life.

The fourth is the move from persuasion to enforcement through institutional capture. Late-stage paradigms increasingly use the state, regulatory bodies, professional associations, and other institutional levers to enforce orthodoxy that would not survive open argument. The capture is what makes the late stage feel oppressive even to people who agree with most of the paradigm’s substantive claims. The argument has been outsourced to enforcement because the argument can no longer carry its own weight.

The fifth, and the most important for predicting what happens next, is that the escalation accelerates collapse. Each high-profile excommunication produces some converts to the orthodox position and a larger number of quiet dissidents who now know the paradigm is fragile. Late-stage enforcement is visible to observers, and visibility teaches the lesson that the paradigm cannot defend itself by argument. Once that lesson is learned widely enough, the paradigm is finished, often quite suddenly.

The three historical cases I will examine each demonstrate these dynamics with unusual clarity, and each adds something the others do not.

The Late Qing: A Paradigm That Tried to Reform Itself to Death

In the autumn of 1898, a thirty-year-old reformist scholar named Tan Sitong sat in his Beijing study and refused to flee. His friends had warned him. The Empress Dowager Cixi had launched a coup against the Guangxu Emperor three days earlier, the Hundred Days’ Reform was over, and the reformers who had advised the young emperor were being rounded up. Most of Tan’s colleagues escaped to Japan with the help of Japanese diplomats. Tan stayed.

“Reform has never come about in any country without the flow of blood,” he is recorded as saying. “No one in China in modern times has sacrificed himself for the cause of reform, and because of this China is still a poor and backward country. Therefore, I request that the sacrifices begin with myself.”

He was executed at Caishikou marketplace on September 28, 1898, along with five other reformers. They became known as the Six Gentlemen of the Hundred Days. The crowd that gathered to watch was reportedly subdued, uncertain what the executions meant. The Confucian-imperial paradigm that had organized Chinese political life for two and a half centuries had just publicly killed its own attempts at self-reform.

The Qing dynasty had been under increasing pressure since the First Opium War of 1839-1842. By the 1890s the cumulative humiliations had become impossible to absorb within the existing paradigm. The Treaty of Shimonoseki in 1895, which ceded Taiwan to Japan and acknowledged Japanese suzerainty over Korea, was particularly devastating because it represented defeat at the hands of an Asian power that had been considered tributary just decades before. The Guangxu Emperor, then twenty-seven years old, had been persuaded by reformers including Kang Youwei and Liang Qichao that the dynasty could only survive through rapid modernization on the Meiji model.

The Hundred Days’ Reform launched in June 1898 attempted to do in three months what Meiji Japan had taken twenty years to accomplish. New schools were to be established. The examination system was to be revised. The military was to be reorganized. The bureaucracy was to be streamlined. The reforms were both too ambitious and too poorly implemented, and they alarmed the conservative faction at court led by the Empress Dowager Cixi. On September 21, Cixi placed the Guangxu Emperor under house arrest, where he would remain for the rest of his life. Most of the reforms were reversed within weeks.

This is the first move of paradigm defense. The conservative faction had defeated the reformers. The paradigm appeared safe.

Two years later the same conservative faction backed the Boxer Rebellion, a violent anti-foreign movement that promised to drive Western influence out of China through martial-arts mysticism and popular mobilization. The Empress Dowager declared war on the foreign powers in June 1900. The Boxers besieged the foreign legations in Beijing for fifty-five days. The Eight-Nation Alliance assembled a relief force, fought its way to Beijing, and occupied the city. The Empress Dowager fled to Xi’an in disguise. The Boxer Protocol of 1901 imposed indemnities equivalent to roughly seventeen times the Qing government’s annual revenue and stationed foreign troops in the country.

This is the second move. The defense that was supposed to shore up the paradigm produced the worst foreign humiliation in Chinese history. The conservatives had now delegitimized themselves with the educated Chinese elite, with the foreign powers, and with their own provincial officials, who had refused to follow the declaration of war and had negotiated separately with the foreign powers to maintain regional stability.

The Late Qing reforms followed from 1901 to 1911. The court, having tried suppression and then reactionary mobilization, now attempted the modernization it had executed reformers for proposing three years earlier. The reforms were comprehensive and desperate. The Imperial Examinations were abolished in 1905, which severed the central recruitment mechanism for the paradigm’s bureaucratic class and broke the linkage between elite Chinese family investment in Confucian education and political opportunity. Constitutional reform was attempted, with provincial assemblies established in 1909 and a national assembly in 1910. A modern military was built, including the Beiyang Army that would later produce most of the warlords of the Republican period.

This is the third move, and it failed for the reason the framework predicts. The reforms removed the structural mechanisms that had sustained the paradigm without successfully replacing them with mechanisms that could sustain a successor paradigm. The educated elite that would have staffed the new constitutional system had been raised under the old examination system and had built their careers around its content. The new military was loyal to its commanders rather than to the dynasty. The provincial assemblies became platforms for elite criticism of the imperial court rather than instruments of its modernization.

The Wuchang Uprising of October 10, 1911 began as a barracks mutiny by soldiers of the New Army in Hubei province. Within two months, fifteen of the eighteen provinces had declared independence from the Qing. The six-year-old emperor Puyi abdicated on February 12, 1912. The dynasty that had ruled China for two hundred sixty-eight years dissolved without a single major battle being fought in its defense.

The paradigm had tried to defend itself through suppression, through reactionary mobilization, and through controlled modernization. Each defense alienated a constituency that might have supported the paradigm. By 1911, no constituency remained. The defenders had spent fourteen years eliminating every group that might have fought for them.

Meiji and the Last Samurai

Thirty-five years earlier and seven hundred miles to the east, a different paradigm-defense episode had played out with different results.

In the predawn hours of September 24, 1877, on a hill called Shiroyama overlooking the city of Kagoshima, a forty-nine-year-old man named Saigō Takamori sat in the entrance of a cave dressed in the formal court robes of his rank and watched the Imperial Army artillery zero in on his position. He had perhaps three hundred men left, from a force that had once numbered thirty thousand. He had been wounded in the hip by gunfire. He could no longer walk. His last followers were preparing to charge down the hill into the Imperial lines, which would be the last cavalry charge of the samurai era.

Saigō asked his closest retainer, a man named Beppu Shinsuke, to behead him. Beppu performed the act with a single stroke, then led the remaining samurai down the hill to die. By 7 a.m. the rebellion was over. The body count for the day was approximately five hundred samurai. The Imperial Army losses were under a hundred.

Saigō Takamori had been, less than a decade earlier, one of the architects of the Meiji Restoration that had now killed him. He had personally commanded the Imperial forces that defeated the last Tokugawa loyalists in 1869. He had served as one of the principal members of the new Meiji government, helped draft its foundational policies, and represented the regime during the Iwakura Mission years. His face appears on Japanese banknotes today. The Imperial Army that killed him had been built largely on his initiative.

The samurai paradigm that Saigō ultimately died defending had organized Japanese society for two and a half centuries under the Tokugawa shogunate. Samurai legitimacy rested on military function, hereditary status, and a Confucian ethical content that elevated bushido as a way of life rather than an occupation. By 1868, when the Meiji Restoration brought the young Emperor Meiji nominally to power, the samurai paradigm was clearly inadequate to confront Western military and economic pressure. The Meiji oligarchy, drawn substantially from samurai backgrounds in the Satsuma and Chōshū domains, set about dismantling the paradigm that had produced them.

The dismantling was rapid and comprehensive. The hereditary stipends that had supported the samurai class were converted to interest-bearing bonds in 1873 and then phased out. The abolition of the han system in 1871 eliminated the political-administrative units within which samurai had held office. The conscription law of 1873 created a national army that broke the samurai military monopoly. The Haitōrei Edict of 1876 banned samurai from publicly wearing swords, the visible badge of their status. Within eight years of the Restoration, the entire structural apparatus that had sustained the samurai paradigm had been demolished by the same coalition that had emerged from it.

The defense came from samurai who could not accept the dissolution. Several smaller rebellions occurred between 1874 and 1877. The Satsuma Rebellion under Saigō was the culmination. Saigō had resigned from the Meiji government in 1873 over the rejection of his proposed Korean expedition, which would have provided military employment for the displaced samurai class. He retired to Kagoshima and established private military academies that became gathering places for samurai discontent. The Meiji government, watching the academies grow into something that looked increasingly like a rival military force, attempted in early 1877 to remove arms from the Kagoshima arsenal. The local samurai rose under Saigō’s leadership.

The rebellion lasted nine months and was fought with a brutal asymmetry that the samurai paradigm had not anticipated. The Imperial Army, built on Saigō’s earlier initiative, was equipped with modern rifles, artillery, and telegraph communications. It used conscript infantry organized in Western tactical formations. The samurai forces relied on sword charges and traditional formations that had been the operative military doctrine for three centuries. At the Battle of Tabaruzaka in March 1877, the heaviest engagement of the war, the Imperial Army’s superior firepower destroyed the samurai assault despite the rebels’ superior individual training. The war was settled at Tabaruzaka. The remaining six months were a long retreat south through the Kyushu mountains to Saigō’s final stand at Shiroyama.

The case illustrates a different paradigm-defense pattern than the late Qing. The Meiji oligarchy had succeeded in installing a successor paradigm before the old paradigm’s defenders could mobilize. By the time Saigō raised his banner in 1877, the new conscript army existed, the new bureaucracy was functioning, the new educational system was producing students who would not be samurai, and the new economic structure was building the industrial base that would soon make Japan an East Asian power. The samurai defense came too late, against an opponent that had already established structural superiority. The defense was contained, brief, and final.

The contrast with the late Qing is the analytic payoff. Both cases involve East Asian states confronting Western pressure in the late nineteenth century. Both involve paradigm transitions managed by elites who recognized the existing paradigm could not survive. The Meiji case produced a controlled transition with a violent but contained final defense by a portion of the old paradigm’s defenders. The late Qing case produced an uncontrolled collapse after fourteen years of failed defensive moves that alienated every potential supporting constituency.

The same population could have had either outcome. The difference was in the elite handling of the paradigm-defense phase. Japan’s managed transition produced a rapid modernization that culminated in catastrophic war by 1945 and rebuild after. China’s mismanaged transition produced the warlord period, the Nationalist-Communist civil war, the Japanese invasion, and eventually the Cultural Revolution. Tens of millions of people died in the difference.

Lysenkoism and the Lethal Science

The third case shows the same architecture operating in a completely different cultural and political context.

In August 1948, at the conclusion of a now-famous session of the Lenin All-Union Academy of Agricultural Sciences in Moscow, an agronomist named Trofim Lysenko announced that his views on heredity had been approved by the Central Committee of the Communist Party. The implication was unambiguous. Mendelian genetics, the operative framework for biological science globally, was now officially counter-revolutionary in the Soviet Union. Soviet biologists who continued to teach or research Mendelian genetics would face consequences. Within weeks, three thousand biologists had been dismissed from their positions. Within months, the entire field of Soviet genetics had been brought under Lysenko’s control or driven into hiding.

Lysenko had been promoting his alternative theory of inheritance, based on a misreading of Lamarckian acquired characteristics, since the early 1930s. His theory had several political advantages in the Stalinist context. It promised rapid agricultural improvements through environmental manipulation, which fit the Five Year Plan ideology of remaking nature through revolutionary effort. It treated heredity as malleable, which paralleled the Marxist-Leninist view of human nature as malleable. And it was associated with peasant practical knowledge rather than with bourgeois Western science, which fit the broader anti-cosmopolitan campaigns of the late Stalin period.

The empirical case for Mendelian genetics had been settled globally by the 1940s. Lysenko’s experiments did not replicate, his crop yields were fabricated, and his theoretical claims were inconsistent with the rapidly expanding body of genetic research being conducted in the United States, Britain, and elsewhere. The Soviet defense intensified anyway.

Nikolai Vavilov, the leading Soviet geneticist of his generation and one of the most distinguished biologists of the twentieth century, had been arrested in 1940 after years of conflict with Lysenko. He had built the world’s largest seed bank at the Institute of Plant Industry in Leningrad, traveled to five continents collecting plant specimens, and developed the theory of centers of origin that remains foundational in modern crop science. He was sentenced to death in 1941, the sentence was commuted, and he died of starvation in Saratov Prison in January 1943.

The Vavilov case is worth pausing on because it shows what paradigm defense looks like at the individual level when the institutional apparatus available for enforcement includes secret police and prison camps. Vavilov had been internationally famous. He had been elected to foreign academies. He had been received by foreign heads of state. None of it mattered once Lysenko had captured the Soviet biological establishment. The defense of the paradigm was sufficient justification for killing one of the most accomplished scientists of the era through neglect in a Soviet prison.

The paradigm held until Khrushchev’s fall in 1964 and was not formally repudiated until the late 1960s. By that point, Soviet biology was a generation behind the rest of the world. The country that had pioneered seed collection and crop science under Vavilov had become incapable of conducting basic genetic research. The enforcement that was meant to protect the paradigm had destroyed the institutional capacity that would have been needed for any transition to a successor paradigm.

When Lysenkoism collapsed, there was no Soviet genetics community ready to take over. There were only the surviving scientists who had spent twenty years hiding their actual views, and the next generation of students who had been taught a framework they were now expected to abandon.

The pattern is recognizable from the late Qing case. The defenders of a failing paradigm sometimes destroy the institutional infrastructure that would have been needed to manage transition to a successor paradigm. They prevent change by destroying the capacity for change. When the paradigm collapses anyway, as it eventually must, the collapse is worse than it would have been if managed earlier.

The Red Guards, Revisited

Different cultures, different centuries, same architecture. Three cases ranging from Confucian imperial bureaucracy through samurai military aristocracy to Soviet scientific orthodoxy, each showing the same five-stage defensive pattern, each collapsing in ways the defenders could not control.

The Cultural Revolution is the case that brings the framework to its limit, because it shows what happens when the paradigm-defense apparatus is turned outward toward the broader society rather than inward toward institutional adversaries. The dynamics are the same. The scale is different. The Lysenko purges killed thousands of scientists. The Cultural Revolution displaced or killed millions of people across every sector of Chinese society.

But the question that opens this piece is not how many died. It is who did the killing, and why.

The standard narrative on the Red Guards is ideological fervor. Mao called, and the youth answered. This narrative is comforting because it makes the Red Guards into a unique phenomenon, a one-time eruption of revolutionary madness that has nothing to do with normal human behavior under normal conditions. If the Red Guards were uniquely crazy, then the Cultural Revolution is a closed case. It does not have lessons for the present.

The historical record tells a different story. Andrew Walder’s research on Red Guard recruitment patterns, drawn from school records and contemporary accounts, shows that the early Red Guards were predominantly the children of Communist Party cadres and military officers. They were specifically the youth whose structural privilege was most directly threatened by the pragmatic reforms of 1962-1965, when the recovery from the Great Leap Forward had begun rewarding competence over class background.

Those reforms had begun opening university admissions to students from intellectual and bourgeois families because the country needed engineers and scientists. The cadre children, raised on revolutionary mythology and the expectation of preferred treatment, could watch their guaranteed advantage eroding in real time. When Mao offered them the chance to defend the revolution against revisionism, he was offering them a mechanism to halt and reverse a process that was disadvantaging them specifically.

This is the first incentive. The defense of the revolution was simultaneously the defense of the class-status system that had given them their advantage.

The career incentive was the second. Universities had closed by mid-1966. Normal career paths had stopped functioning. In their place, Red Guard leadership offered a new kind of career. Prominence in struggle sessions. Visibility in mass rallies. Organizational responsibility within factions that controlled access to denounced households, party records, and confiscated property. The chance to be received personally by Mao at the Tiananmen rallies of August through November 1966, an experience that became a credential deployable in later contexts. An estimated thirteen million Red Guards traveled to Beijing during this period.

For a teenager from a provincial city, the trip to Beijing to see the Chairman was the most important event of their lives to that point, and they understood at the time that it would matter for the rest of their lives.

The factional dynamics produced a third incentive. By 1967, the Red Guard movement had splintered into competing factions. Faction membership produced both protection and threat. Being in the wrong faction at the wrong time could result in physical violence from rival Red Guards. The factional competition turned the movement into a hierarchical organization with enforced loyalty, where individual participants had strong incentives to escalate their commitment to their own faction in order to secure their own safety.

The safety incentive was the fourth, and may be the most underrated. By late 1966, not participating in Red Guard activity was itself a marker of suspicion. Students who declined to denounce teachers, refused to participate in struggle sessions, or expressed discomfort with the violence were liable to be denounced themselves. The choice was between participating actively, participating passively, or being targeted. Active participation provided protection. Passive participation invited scrutiny. Refusal invited destruction.

Many Red Guards described, in memoirs written after 1976, that their participation was driven substantially by fear of being targeted if they did not participate enthusiastically enough. This is the dynamic that produces active complicity in every paradigm-defense episode. The pressure to be visibly enthusiastic about enforcement as a form of insurance against becoming its target.

The identity incentive was the fifth. The Red Guard role offered Chinese teenagers a sense of historical importance that was not available through any other channel. They were not ordinary students. They were the vanguard of world revolution, personally entrusted by Mao with the task of saving the country from revisionism. The psychological power of this framing for adolescents who had grown up on revolutionary mythology cannot be overstated. The role transformed teenage anxieties about future and status into world-historical mission. The transformation was psychologically real, and it bound participants to the movement through identity stakes that operated independently of the material rewards.

When you add these five incentive structures together, you have a system that produced approximately what any such system would produce in any human population. Not because Chinese teenagers were uniquely susceptible to ideological madness, but because the incentive structures were aligned with paradigm enforcement, and human beings respond to aligned incentive structures with reasonable consistency.

Ideology supplied the justification and the coordination signal. Material interests, career incentives, status incentives, identity incentives, and safety incentives supplied the actual energy. This is the careerist reading of the Red Guards, and it is much darker than the madness reading because it implies that the conditions that produced Red Guard behavior are not specific to Maoist China. They are conditions that emerge whenever a paradigm under threat aligns enough incentive structures behind its defense.

Then the incentive structure shifted.

By 1968, Mao had decided that the Red Guards had served their purpose and were now a source of disorder. The Down to the Countryside Movement was launched at the end of 1968 and ran through 1980, sending an estimated seventeen million urban youth to rural areas for indefinite agricultural labor. The structural privilege they had mobilized to defend ended up destroyed in the process. The class-status system had been so disrupted by the chaos they helped create that it could not be restored to its pre-1966 form.

The true believers were devastated. The opportunists repositioned. The lost generation became the cohort most receptive to Deng’s market reforms in the late 1970s, because the political route to advancement had burned them once and they were not going back.

The Reformed Emperor

There is a figure who haunts the Cultural Revolution from inside the regime’s own success stories, and his case complicates everything the careerist framing of the Red Guards makes too clean.

In 1959, a fifty-three-year-old man named Aisin-Gioro Puyi walked out of the Fushun War Criminals Management Centre in northeastern China. He had spent the previous ten years there, undergoing systematic ideological reformation under a program the Chinese state had developed during the Yan’an period and refined for use on captured Nationalist officers, Japanese war criminals, and ideologically suspect intellectuals.

He had arrived at Fushun unable to dress himself or tie his own shoes, having been waited on by servants his entire life. The administrators of the program understood that personal helplessness and ideological pride were structurally connected, and that breaking the former was a precondition for restructuring the latter. By 1959 he could tie his shoes, raise vegetables in the prison garden, and produce extended written confessions of his past errors in the appropriate Marxist-Leninist vocabulary.

Puyi was the last emperor of China. He had been enthroned at age two in 1908, deposed at six in 1912 during the revolution that ended the Qing dynasty discussed earlier in this piece, restored briefly in 1917, installed as the puppet emperor of the Japanese-controlled state of Manchukuo from 1934 to 1945, captured by the Soviets at the end of the war, and handed over to the Chinese Communists in 1950. The years at Fushun were the longest period of stability his life had known.

After his release he worked as a gardener at the Beijing Botanical Gardens, married a hospital nurse named Li Shuxian in 1962, and wrote an autobiography titled From Emperor to Citizen that was published in 1964. The book is one of the most extraordinary documents of the Mao era. It describes his life as a sequence of progressively deepening errors that ended in genuine moral awakening through the patient reformation provided by the Communist Party. The framing is the official framing. The vocabulary is the official vocabulary. The conclusions are the official conclusions.

The question is whether Puyi believed what he wrote.

The honest answer, the one the historical record supports, is that he believed some portion of it and performed the rest, in proportions that no one, including Puyi himself, could ever cleanly separate. He reported, in private conversations with relatives and in his autobiography, that he had come to see his previous life as deluded, his collaboration with the Japanese as shameful, and the new China as a legitimate moral project that had improved the lives of the Chinese people. His half-brother Pu Jie, who underwent the same reformation, made similar reports independently.

The reform was not pure theater. Something in his actual self-understanding had been reorganized.

The performed element was also real. He knew exactly what his survival required. He produced the expected confessions and denounced his own past with the vocabulary the administrators provided. The performance was a survival strategy and he understood it as one. The opportunism was structurally necessary and he was not in a position to refuse it. His status as a successfully reformed emperor was useful to the state, which paraded him at official functions as proof that the program worked. Continued performance of the reformed identity was rewarded continuously throughout the rest of his life.

This is the part of paradigm-defense analysis that the careerist framing of the Red Guards does not fully reach. The opportunist-versus-true-believer binary is analytically useful but psychologically thin. Real reform produces real changes in self-understanding that are not reducible to performance. Real performance produces effects on belief that are not reducible to opportunism. Puyi sat inside the structural ambiguity for the rest of his life, and the ambiguity was not a moral failing on his part. It was the actual condition of being a human being subjected to sustained ideological pressure by a state that had absolute power over his physical existence.

By the time the Cultural Revolution began in 1966, Puyi was a publicly model reformed citizen. Married, employed, writing supportive editorials about the revolutionary process. His former servants and minor relatives were being denounced as feudal remnants. His autobiography was being criticized as insufficiently self-critical. He himself was protected from the worst of it by Zhou Enlai’s direct intervention, but the protection was provisional.

He died of kidney cancer on October 17, 1967, having been refused proper medical care at the hospital because of his class background until Zhou once again personally intervened. But the intervention came too late to save him. The reformation that had restructured his identity over a decade did not protect him from being killed, slowly and bureaucratically, by the paradigm whose triumph he had been used to demonstrate.

The Red Guards who attacked his physicians in 1967 included people who had been children when his reformation was completed. They had grown up on the official story that the program worked, that even the last emperor could become a citizen, that the new China could absorb and transform anyone. They were now killing the people whose absorption had been used as evidence for the absorption’s possibility.

The paradigm was eating its own demonstration cases. The model workers, the rehabilitated landlords, the reformed intellectuals, the Soviet-trained engineers who had been publicly praised in the 1950s were all liable to be denounced in the 1960s by the children of the people who had once praised them. The system did not honor its own success stories. It used them while they were useful and discarded them when the next phase of enforcement required new targets.

There is a non-cynical observation worth making about this.

Some participants in any paradigm-defense episode are genuinely transformed by it, in ways that are real even when the transformation includes opportunistic elements. Those participants tend to be the ones who suffer most when the paradigm shifts again, because they cannot smoothly reposition. The opportunists who never converted have nothing to lose by re-converting to the next paradigm. The true believers who actually reformed have lost the identity they spent years building.

This is a recurring feature of paradigm-defense episodes that should produce more compassion for true believers than they typically receive, even from people who disagree with what they believed. The true believer is the one paying the highest price for a coordination device that the opportunists were using and discarding throughout.

Puyi was a true believer, in the limited sense that any of his belief was real, and his belief did not save him. The paradigm killed him at the moment of its greatest apparent triumph, and the killing was bureaucratic, slow, and absolutely indifferent to the conversion the paradigm had spent ten years producing in him.

Preference Falsification and the Speed of Collapse

There is a concept from the economist Timur Kuran that helps explain why paradigm collapses feel so sudden when they finally arrive: Preference falsification.

The idea is very straightforward. In environments where holding a particular view carries social or professional cost, people will publicly profess beliefs they do not privately hold. Each individual’s public conformity reinforces the perception that the paradigm is universally supported, which increases the pressure on others to conform, which produces more false signals of support. The system appears stable, but the gap between public performance and private belief widens steadily with each enforcement cycle.

The late Soviet Union is the cleanest illustration. By the mid-1980s, the official ideology had been privately abandoned by the vast majority of the population, including many of the people whose jobs required them to enforce it. But each individual assumed that everyone else might still believe, because the public signals all pointed toward belief. Party functionaries gave the speeches. Citizens attended the rallies. Institutions performed the rituals. The maintenance of public conformity sustained the appearance of consensus that did not actually exist.

When Gorbachev’s reforms cracked the enforcement apparatus just enough to demonstrate that dissent was survivable, the cascade was extraordinarily fast. The population in 1989 was discovering, all at once, that everyone else’s mind had already changed years ago. The speed of the collapse was proportional to the gap between the public performance and the private reality. The wider the gap, the faster the cascade, because there was less genuine conviction to overcome.

This explains why paradigm-defense episodes that rely on enforcement after persuasion has failed are structurally fragile. Every act of enforcement that produces public compliance without private conversion widens the gap between surface and substrate. The paradigm looks stronger with each successful enforcement action, even as it becomes more brittle, because each coerced conformist is a future defector waiting for the signal that defection is safe. The signal, when it comes, can be remarkably small relative to the cascade it triggers.

A few prominent defections. A shift in institutional leadership. A lost election. A policy reversal at a single major organization. Any of these can serve as the coordination signal that tells the falsifiers that others are moving and they can move too.

The contemporary West has just lived through a compressed version of this cycle. Positions that were mainstream in 2012 became professionally dangerous by 2020, and positions that were professionally dangerous in 2020 are voiced freely in 2026, sometimes by the same institutions that had been enforcing against them. The speed of the reversal suggests a preference-falsification gap that was wider than the enforcers understood.

But the more interesting question, the one the historical cases all point toward, is what comes after.

The Interregnum

What follows a paradigm collapse is an interregnum, borrowing from Gramsci’s line that the old world is dying and the new world struggles to be born. The old paradigm has lost enforcement capacity, but no replacement has consolidated. Former enforcers are still in position, still partially aware their authority has evaporated, unable to bring themselves to acknowledge it publicly.

Historical interregnums share features very much worth recognizing.

Post-Mao China took roughly fifteen years to consolidate a new operating framework under Deng. The post-Soviet settlement in Russia took most of the 1990s to produce a recognizable successor regime, and arguably is still being contested in the form that emerged under Putin. Meiji Japan took twenty years after the Satsuma Rebellion to consolidate the modernization paradigm that would carry the country into the twentieth century. The enforcement collapses quickly. The replacement settlement takes much longer.

Former enforcers do not generally repent or convert. They mostly do one of three things. Performative continuation, where they keep saying the things they used to say in the same forums, increasingly to audiences that no longer respond, until they retire or are replaced. Silent repositioning, where they stop saying the controversial things, gradually shift their language into more defensible territory, and pretend their previous positions had always been more nuanced than they actually were. Or genuine public reckoning, where someone writes the essay or book that says, this is what I did, this is what I now think I got wrong, this is what I have learned. The third group is small but disproportionately influential because their writing tends to set the terms by which the period is later understood. The first two groups produce no durable contribution to the new synthesis.

Both sides typically miscalibrate the risks. The optimistic side assumes the new settlement will be the natural restoration of pre-collapse conditions. This almost never happens. Post-Reformation Europe was not a restoration of pre-Reformation Catholicism. The post-Soviet settlement in Russia was not a restoration of pre-Soviet conditions. The pessimistic side predicts catastrophic regression. This also rarely materializes. What emerges is usually a messy contest of partial frames, most institutional life continuing under less ideological pressure in a muddled middle that satisfies nobody completely.

The Descriptive Position

The implication of all of this, if you are trying to write or think clearly during a paradigm transition, is that the costs of explicit dissent will be high right up until they are not, and the transition will be sudden. People treated as beyond the pale for views that recently qualified as mainstream will find themselves treated as obvious within five to ten years, sometimes (often, even) by the same people who were enforcing the orthodoxy against them.

This is an observation about structure, about what paradigm shifts look like in real time, regardless of which paradigm is dying. The enforcers and the enforced often end up on the same side of the new paradigm, and the enforcers usually do not acknowledge that they switched.

The writing that matters in these moments, the writing that will still be cited when historians try to understand what happened, is writing that describes the transition’s structure without partisanship for either the dying paradigm or its opposition. The defenders of a collapsing paradigm cannot write its honest history because the history indicts their enforcement. The opponents can write polemic but not analysis. The descriptive observer who treats the period as a case study inside a longer historical sequence is the one whose work survives.

The defensive escalation itself is the tell. Confident paradigms absorb dissent. Paradigms in their late stage cannot afford to, and the aggression with which they pursue dissenters marks the phase of the cycle more reliably than any other signal.

When you see it, you know the paradigm is dying, and the work that will endure is the work that describes the late stage as the late stage rather than fighting within it.

The teenagers in the Beijing courtyard in August 1966 were not crazy in the traditional sense. They were sixteen years old and they were responding to incentive structures that the paradigm’s defense had constructed around them. The vice principal they killed had been their teacher. The vocabulary they used to denounce her had been supplied. The protection they sought through enforcement was real, and so was the danger if they declined to enforce. They acted reasonably, in a structural sense, and their reasonable action killed a fifty-year-old woman who had four children and had taught at their school for seventeen years.

This is what paradigm defense looks like at its worst. The participants experience themselves as historically necessary. The structure that produced them is invisible to them from inside. The structure becomes visible only after the paradigm collapses, and by then the participants have either repositioned, gone silent, or been destroyed.

I intend to keep describing the structure while it is still operating. If the work is useful to you, pass it along.

On May 17, 1994, a group of South Korean government scientists walked into the president’s office and presented a memo about dinosaurs. Not real ones. Steven Spielberg’s dinosaurs.

Their argument was simple: Jurassic Park had grossed $850 million worldwide, which was equivalent to the foreign sales of 1.5 million Hyundai automobiles. Korea, at the time, exported barely 700,000 vehicles per year.

A single Hollywood film was worth more than twice the country’s entire annual car export volume.

The president read the memo. The government created a Cultural Industry Bureau.

Four years later, in the middle of the worst financial crisis in the country’s history, South Korea’s new president made a bet that most economists thought was insane: he would spend his way out of the crisis by manufacturing pop culture.

Thirty years later, 9 of the 10 best-selling albums on the planet are Korean.

This is the story of how that happened.

The Memo

The memo wasn’t vague. It wasn’t a soft plea about “investing in the arts.” South Korea’s Presidential Advisory Council for Science and Technology titled it “Strategic Plan for the Growth of High-Tech Visual Arts,” and the arithmetic was blunt: Jurassic Park grossed $850 million on a $63 million budget.

That revenue equaled the foreign sales profit of 1.5 million Hyundai cars. Korea exported barely 700,000 vehicles that year.

A dinosaur movie, made in eighteen months, out-earned the entire country’s annual car export volume by more than two to one.

Scholars Vincenzo Cichelli and Sylvie Octobre later named this “the Jurassic Park syndrome”: the moment popular culture became viewed as a commodity with exchange value exceeding industrial goods.

The government created a Cultural Industry Bureau with an initial budget of 5.4 billion won ($4.1 million). Modest. Almost symbolic.

Then the 1997 Asian financial crisis hit.

South Korea took a $58 billion IMF bailout, the largest at the time. The economy cratered.

And incoming President Kim Dae-jung, inaugurated in 1998, made what looked to most economists like a reckless bet: spend your way out of a financial crisis by manufacturing pop culture.

The policy mechanics were aggressive, to say the least.

Government spending on culture jumped from $14 million in 1998 to $84 million by 2001. The music industry budget alone surged from 940 million won in 1997 to 14.5 billion won in 2000, a fifteen-fold increase in three years.

The 1999 Framework Act for the Promotion of Cultural Industries established the first comprehensive legal definition of cultural products and mandated tax incentives and export subsidies.

The Korea Creative Content Agency (KOCCA) received a mandate to subsidize record companies, fund overseas tours, and staff international music expos, eventually opening offices in the US, France, China, Japan, India, Indonesia, and Vietnam.

Government officials arranged ad sales for Asian TV stations so those stations would accept ready-to-play packages of Korean soap operas, already translated and subtitled, at no cost.

As journalist Euny Hong put it: “There was zero market for Korean popular culture. It had to be force-fed into foreign markets.”

The term “Korean Wave” (Hallyu) was coined by Chinese journalists in a 1999 Beijing Youth Daily article. They expected it to be a fad.

A country in the middle of the worst financial crisis in its history had looked at its balance sheet and decided the answer was boy bands and soap operas.

The Trust Layer

Why would anyone outside Korea care about Korean pop culture in the first place? The sequencing tells you everything. Korean industrial products established global credibility first. Cultural exports rode on that trust.

In the early 1990s, Samsung was known among expats as “Samsuck.” Hyundai’s Excel, launched in 1986 at $4,995 (the cheapest car available in America), was a recurring late-night punchline. Jay Leno made it a regular target. Consumer Reports relentlessly attacked its reliability.

Korea, in the Western imagination, meant cheap goods.

The transformation was very deliberate.

In 1993, Samsung’s chairman Lee Kun-hee gathered 200 executives in Frankfurt and told them to “change everything except your wife and children.”

The company scrapped $50 million in inventory, tripled R&D to $6.5 billion annually, and shifted from lifetime employment to merit-based promotions.

By 2005, Samsung surpassed Sony in global brand value. A Korean brand eclipsing Japan’s most iconic electronics company. By 2011, it overtook Nokia as the world’s largest smartphone maker, a position it held for 14 consecutive years.

In 1998, Hyundai introduced a 10-year, 100,000-mile powertrain warranty. The message: we now believe in our own cars. The brand went from punchline to J.D. Power’s Highest in Initial Quality award by 2019.

By the time Korean pop culture arrived in Western markets in the 2010s, “Korea” no longer connoted cheap goods. It connoted Samsung phones, Hyundai reliability, and LG appliances.

The top 10 Korean conglomerates account for roughly 60% of the country’s GDP.

I visited Seoul in 2024, and what struck me first was the grooming. Everyone was immaculately polished, in a way that went beyond fashion and into something closer to national discipline. I found myself dressing more carefully just to avoid feeling conspicuous.

That sensibility comes from the same coordination culture that turned Samsung from a punchline into the world’s largest smartphone maker: the instinct to engineer quality into every visible surface.

The pop music system just happens to be where that instinct became most legible to the outside world, second only to the dramas.

The Engineer

The government built the policy infrastructure: subsidies, legal frameworks, export support, and overseas offices.

But policy alone does not produce pop stars. Someone had to build the production system that could absorb those resources and convert them into exportable culture at scale.

That person was Lee Soo-man.

Lee was a modestly successful singer and TV emcee in 1970s Korea before leaving for California to study computer engineering at Cal State Northridge. While there, he witnessed Michael Jackson’s rise and the MTV revolution.

He came back with an inversion that reshaped an entire industry:

“Up until that time, the majority of people thought that it is only when you become a big and wealthy country that your culture gets to shine globally. Economy comes first and then culture will follow. But I started thinking in the opposite way, that culture should be first and then economy next.”

He established SM Studio in 1989 in the Apgujeong neighborhood of Gangnam, Seoul. His first signed artist, hip-hop singer Hyun Jin-young, was arrested for drug use, destroying the investment.

That failure was absolutely pivotal. Lee decided to build a system that ensured nothing was left to chance. The scandal inoculated SM against ever trusting talent development to organic processes again.

Modern K-pop, in its entirety, exists downstream of this moment. If you are a fan of BTS, aespa, or Blackpink, you have Mr. Jin-young’s behavior to thank for it.

Around 1997, he coined the term “Cultural Technology” (문화기술) and wrote a literal physical manual that SM employees were required to study.

His head of A&R, Chris Lee, explained to Billboard:

“Mr. Soo-man Lee has this mindset that when it comes down to technology or computer science, there’s always a manual. So he wanted to make sure that we have those manuals, specific manuals to follow to aim for perfection when creating creative content.”

His training as an engineer produced an entertainment system with the rigor of an engineering specification. SM receives approximately 300,000 applicants across nine countries every year. Roughly 100 receive training.

Of all trainees who enter the system, about 1% ever debut. The cost of developing a single idol was estimated at $3 million per member. Debuting an entire group costs $7.5 million or more.

Each group was designed for a specific strategic purpose.

H.O.T. (1996) was formed by polling high school students to determine what teenagers wanted.

BoA (2000) was deployed to conquer the Japanese market at age 13.

Later groups pushed the logic further: sub-units singing identical songs in Korean and Mandarin, city-based franchises designed with unlimited members.

This is what it looks like when government cultural investment meets an engineer’s instinct for systematization: not one artist hoping for a breakthrough, but a production system calibrated to generate them in sequence.

What happens inside that system has a texture that’s difficult to convey without gruesome specifics.

Former SM trainee “Euodias” described it in a 2020 BBC feature: each trainee was assigned a number, and instructors called them only by their numbers, read from stickers on their shirts.

“It felt weird, a bit like we were in some sort of science experiment.”

Trainees were divided into Team A and Team B, “largely based on visual potential and body weight, not skill.”

Weight surveillance is institutionalized. Companies weigh trainees publicly, announcing numbers out loud. Public shaming being insufficient, those over the company’s target are placed on water diets or denied food outright.

Companies cover all training costs as an advance loan; trainees who leave without debuting face penalties ranging from $86,200 to $129,000.

The system reformed somewhat after the TVXQ lawsuit of 2009, when three members of SM’s biggest boy group sued over their 13-year contract, claiming they worked with only four hours of sleep and received almost none of the profits.

The court ruled in their favor, and the Fair Trade Commission limited entertainment contracts to seven years.

The system reformed enough to survive. It didn’t soften.

What the System Produces

The girl group aespa, debuted in 2020, makes a useful case study because the group makes all of this infrastructure visible rather than hiding it.

One member, NingNing, moved from Harbin, China, to South Korea alone at age 14 and trained for five years. She was the only member of her trainee cohort to survive to debut. Every other trainee from her class was cut. Another member, Giselle, trained for approximately eleven months.

Same group. Same debut day.

Their concept pushes the engineering into the aesthetic itself. Each member has a digital AI avatar counterpart existing in a virtual dimension called KWANGYA. Even the members struggled with the lore. But member Karina captured the concept’s tension perfectly:

“æ-Karina doesn’t make mistakes. But I can make mistakes.”

By 2023-2024, SM quietly began retreating from the heavy KWANGYA narrative, shifting toward simpler aesthetics. Even the engineers recognized the concept had outrun its usefulness. But the instinct it expressed, making the construction visible rather than hiding it, remains the group’s defining characteristic.

The Unpaid Operations Department

The government built the policy infrastructure. Lee Soo-man built the production infrastructure. Neither fully explains how Korean music achieved the chart dominance it has today.

That required a third layer: the fans.

K-pop fandoms operate through what Korean fans call chonggong (총공, “collective attack”): coordinated mass action to break digital metrics. This is a decentralized operation performing the work traditionally done by marketing departments, except no one draws a salary.

The sophistication is operational, not just emotional. Streaming guides specify platform rules: on Spotify, play each song for 30+ seconds, switch between 3-4 playlists, include non-K-pop songs to avoid bot detection, pause every 20-30 minutes, mute the device rather than the app.

Fans in Indonesia and India wire money to US-based fans for bulk digital purchases to count toward Billboard charts. BTS ARMY achieved 101.1 million YouTube views in 24 hours for “Dynamite” through organized global shifts.

The Hyundai Research Institute estimated BTS’s annual economic contribution at $4.65 billion, equivalent to the output of nearly 500,000 average Korean workers, or roughly 0.3% of South Korea’s entire GDP.

The scale of fandom coordination has pushed beyond music entirely. BTS ARMY’s #MatchAMillion campaign raised $1,026,531 from 35,609 donors in roughly 24 hours. A single free BTS concert in Busan generated an estimated $500 million in economic value.

And then there’s the chart dominance, and the contradiction at its center.

On the 2024 IFPI Global Album Sales Chart, 9 of the top 10 and 19 of the top 20 best-selling albums worldwide were by K-pop artists. This dominance is partly driven by the fansign lottery system: fans purchase dozens to hundreds of copies of the same album for a chance to meet an idol in person, extract the collectible photocards, and discard the album shells.

Documented photos show boxes of hundreds of K-pop albums dumped on Seoul streets.

Nine of the ten best-selling albums in the world, and piles of them in the gutter.

The Frame

A system like that (numbered trainees, public weigh-ins, six-figure exit penalties) carries a predictable reputation in Western music criticism.

The framing is always the same: manufactured, artificial, factory-made.

But that reaction papers over a philosophical gap worth examining, because it explains why Western critics have such a hard time taking the music seriously, even as it dominates their charts.

Western pop is also extensively engineered. Swedish producer Max Martin has written or co-written over 25 Billboard #1 hits, surpassing the Beatles’ record of 20.

Berry Gordy Jr. explicitly modeled Motown Records on Ford’s assembly line:

“I worked in the Ford factory before the record business and I thought, why can’t we do that for the creative process.”

The machine exists. The cultural convention, inherited from Romanticism’s insistence that real art flows from individual genius, is to obscure it.

A Seoulbeats analysis identified a particularly telling example: New Yorker writer John Seabrook, in 2012, defended the Norwegian production duo Stargate’s role in creating Rihanna’s hits (”She embodies a song in the way an actor inhabits a role, and no one expects the actor to write the script”), then in the same year framed K-pop’s equivalent process as manufacturing in a chapter titled “Factory Girls.”

Same production model. Different framing, depending on who is doing it.

K-pop doesn’t hide its engineering. Instead, it treats visible effort as a selling point.

A 2025 academic paper in The Journal of Creative Behavior drew parallels between K-pop training and the Pansori concept of Gongryeok (功力): accumulated mastery through dedicated practice rooted in Confucian values.

In one framework, engineering is a mark against authenticity. In the other, it’s proof of seriousness.

What makes this tension productive (rather than merely academic) is how different Korean acts have navigated it.

BTS became the biggest act in the world partly by conforming to the Western authenticity frame: all seven members have had songwriting credits since debut, and RM told interviewers,

“We don’t want to change our identity or our genuineness to get the number one.”

They earned credibility within Western criticism by demonstrating personal authorship.

aespa did something closer to the opposite. The AI avatars, the metaverse, and the lore universe are all transparently constructed and even allude to the real-life SM Entertainment within the story.

The group made no bid for the Romantic ideal. They leaned into the engineering and made the engineering itself an artistic statement.

BTS proved Korean artists could meet the Western standard. aespa asked why they should have to.

Culture as Infrastructure

South Korea’s intellectual property exports (music, film, gaming) reached $9.85 billion in 2024, tripling from roughly $3 billion a decade earlier.

Of that total, gaming accounts for 73%; the music and television industries drive disproportionate attention and soft power relative to their direct revenue share.

Global Hallyu fans numbered 225 million in 2023, up from 46 million in 2012. South Korea climbed from 28th to 9th in the Brand Finance Global Soft Power Index for Culture and Heritage between 2020 and 2025, and ranked 1st globally for influential media.

The feedback loop between cultural and industrial exports is now self-reinforcing. K-pop drives interest in Korean cosmetics, fashion, food, tourism, and language study. Cultural exports are building a brand architecture on which everything else appreciates.

Thirty years ago, a government memo compared a Spielberg film to Hyundai sales to argue that culture could be as valuable as industry.

Today, a 14-year-old girl from Harbin moves to Seoul alone, trains for five years, survives a system that cuts everyone around her, and debuts as one quarter of a group whose AI counterparts live in a corporate metaverse at an address that matches the company’s actual building.

The memo’s math has inverted:

Korea no longer measures culture against car exports.

The culture is what makes the world want the cars.

In downtown Bangkok, a 49-story luxury tower stands completely empty. It has been empty for 27 years.

The Sathorn Unique was 80% complete when Thailand’s currency lost half its value in six months, GDP contracted 10.5% in a single year, and 56 of 91 finance companies were permanently shut down.

The developer’s son refused to declare bankruptcy. He wanted to repay every original buyer in full. That never happened. The tower survived the 2025 Myanmar earthquake that shook Bangkok. It became the set of a Thai horror film. It hosted a seminar on the 20th anniversary of the crisis that killed it.

It is, by any measure, the most haunting monument to a financial crisis anywhere in the world. The crisis ended decades ago. The psychology it installed never did.

The first time I went to Bangkok, in August 2023, I couldn’t square what I was seeing with what I thought I knew. The people were visibly sharp. The youth dressed like they’d walked off a Seoul street and spoke English with an ease I hadn’t expected.

One Bangkok, the $3.9 billion mixed-use district designed by the architects behind the Burj Khalifa, was rising in the center of the city. ICONSIAM, a $1.65 billion riverside mega-mall, housed an indoor floating market representing all 77 Thai provinces alongside Louis Vuitton and Chanel.

I’ve been going back regularly since then, and the feeling has only deepened. Thailand is full of people who should be living in a country that works better than it does.

The standard story about developing economies is that the bottleneck is human capital: the workforce is underskilled, the educational pipeline is broken, the culture doesn’t support innovation. Thailand inverts every one of those assumptions and still can’t break through. Between 2015 and 2024, Thailand averaged 1.9% GDP growth. Vietnam posted 7%. Indonesia managed 5%. Malaysia came in between 4 and 5%. Thailand fell from second to fifth among ASEAN economies for foreign direct investment between 2010 and 2022.

Over 3,500 factories shut between 2021 and 2024, with 51,500 workers losing their jobs in the twelve months through June 2024 alone (an 80% year-on-year increase). Subaru ceased all assembly at its Lat Krabang plant after production collapsed to 344 units in four months.

Suzuki announced closure of its Rayong factory by end of 2025. Samsung moved smartphone production to Vietnam years earlier, where its six factories now account for over 50% of the company’s global smartphone output and generate roughly 30% of Samsung’s total revenue. The consumer side is cracking alongside manufacturing: auto loan non-performing loans surged 32% year-on-year, car leasing NPLs hit 27.25% in Q4 2024, and commercial bank lending has contracted for four consecutive quarters, the longest contraction in over 20 years.

The numbers tell you what happened. They don’t tell you why. The answer is a structural trap with interlocking parts: a crisis that installed permanent caution, a concentration of wealth that was strengthened by that caution, and a demographic collapse driven by the resulting lack of opportunity. Each piece makes the others harder to fix, and the combined weight makes the trap politically unsurvivable: no leader lasts long enough to coordinate an escape.

The Psychology That Never Left

On July 2, 1997, the Thai baht was unpegged from the dollar after 14 years at a rate of 25:1. By January 1998, it had plunged to 56. Of the $38.7 billion in foreign reserves, only $2.85 billion remained. The IMF assembled a $17.2 billion rescue package. Over 300 high-rise projects were abandoned mid-construction across Bangkok. U.S. diplomat Marie Therese Huhtala recalled people pulling their kids out of school, executives losing everything, wealthy families selling furs and Mercedes at fire-sale prices.

The Bank of Thailand itself acknowledges the crisis left “remaining scars.” These manifest as institutional risk aversion of an intensity that is difficult to overstate. Foreign reserves were rebuilt from $2.85 billion to over $237 billion by 2018. Banking regulation became among the strictest in the region, with capital adequacy ratios at 17.8% and loan-loss reserves at 162% of required amounts. Corporate investment has been persistently weak since 1997.

Here is the frame that got installed: caution above all else. Safe export-oriented manufacturing. Tourism. Strategies that generate foreign exchange without the vulnerability of foreign borrowing. These were rational choices. They were also the choices of a country that would never again bet on itself the way it had in the late 1980s, when Thailand’s economy was growing at over 9% annually, the highest rate of any country on Earth.

The aspiration was set in that era. The caution was set by what came after. Both are still running simultaneously, and neither can overwrite the other. That tension defines modern Thailand more than any single policy failure.

And the caution didn’t just shape policy. It shaped who won.

Where the Money Sits, and Where It Doesn’t

If the 1997 crisis installed the psychology, the psychology installed the structure.

A banking system that will only lend to proven balance sheets lends to conglomerates. A regulatory system terrified of destabilizing the economy doesn’t challenge incumbents. An investment climate that only funds what’s already validated funds the people who already own everything.

Thailand didn’t develop extreme wealth concentration because a few families seized power. It developed extreme wealth concentration because the caution frame, applied consistently for 27 years, has a natural output: whoever was big enough to survive 1997 absorbed everything that came after.

Credit Suisse’s 2018 Global Wealth Databook ranked Thailand the most unequal country in the world among 40 nations surveyed. The richest 1% controlled 66.9% of all wealth. The bottom 50% held 1.7%. But the numbers matter less than the mechanism.

CP Group was forced to sell Lotus during the 1997 crisis. Then it bought everything back (Makro in 2013, Tesco Lotus for $10.6 billion in 2020) at moments when smaller competitors couldn’t finance operations. Today, CP Group controls 14,000 7-Eleven stores, Thailand’s largest telecom operator, and one of the world’s largest agribusiness operations.

TCC Group controls Thai Beverage, massive real estate holdings, and is the developer behind One Bangkok. Together with Boon Rawd Brewery, TCC controls over 90% of Thailand’s beer market. King Power holds a near-total monopoly on duty-free retail.

The combined wealth of Forbes Thailand’s top 50 reached $170.5 billion in 2025, roughly 33% of the country’s entire GDP.

Thailand’s 1999 Trade Competition Act yielded zero prosecutions in its first 17 years. When the True-DTAC telecom merger reduced the mobile market from three to two major players in 2023, the deal proceeded without resistance. This wasn’t corruption in the traditional sense; it was the caution frame working exactly as designed: don’t disrupt what’s functioning, don’t risk instability, don’t bet on an outcome you can’t guarantee.

Every subsequent downturn repeated the cycle. COVID was the most recent iteration: the big conglomerates were described as “among the few cashed-up Thai buyers” for distressed bank portfolios. The crisis that broke the middle class made the conglomerates stronger. Then the caution that followed the crisis made them permanent.

A Country Aging Without Getting Rich

The detail that reframes everything is the birth rate.

Thailand’s total fertility rate fell to 1.0 in 2024. Lower than Japan’s 1.2. Approaching South Korea’s record-breaking 0.72. But South Korea’s GDP per capita is five times higher, at $36,000 versus Thailand’s $7,300. Thailand is the only Southeast Asian country the UN classifies among nations with declining birth rates, a group that otherwise consists entirely of high-income, developed economies.

In 2024, Thailand recorded 462,240 births versus 571,646 deaths. The first time births fell below 500,000 since 1949. The fourth consecutive year deaths exceeded births. Thailand’s median age of 40.8 matches the United Kingdom’s. The ASEAN average is 29. By 2050, one in three Thais will be over 60, and the working-age population will shrink from 43.2 million to 36.5 million.

This demolishes the standard assumption that birth-rate collapse is purely about wealth. Thailand’s plunge below replacement fertility began in 1991, driven by an extraordinarily successful 1970s family planning program that cut population growth from 3.1% to 0.4%.

But the continued decline, a survey found only 35.8% of fertile Thai adults plan to have children, is driven by something harder to measure: the sense that the future doesn’t hold enough promise to invest in.

The conglomerate structure and the demographic decline reinforce each other: young people opt out because the economy offers them no path forward, and their departure further concentrates the domestic market in the hands of the incumbents who already dominate it.

The “Let’s Move Abroad” Facebook group, created in May 2021, attracted nearly 1.1 million members almost immediately. In a country of 72 million, that is a mass expression of generational despair.

Thailand produces 400,000 to 500,000 university graduates annually, fewer than half of whom find immediate formal employment. Starting salaries hover around 15,000 to 20,000 baht per month ($420 to $560).

A single BTS/MRT fare of 39.50 baht represents 11% of the daily minimum wage, the highest proportion in Asia. Bangkok earned UNESCO Creative City of Design designation in 2019. The 2024 film How to Make Millions Before Grandma Dies was an international breakout. The creative talent is unmistakable. The economic infrastructure to support it is not.

Breaking free would require coordinated industrial policy sustained over decades. An economy this rigid, a demographic clock this advanced, a debt burden this heavy: the structural conditions make that coordination almost impossible.

Thailand has had 11 prime ministers in 20 years, not because its politics are uniquely dysfunctional in isolation, but because the trap itself is politically unsurvivable. No leader holds power long enough to execute a strategy that takes a generation to pay off.

Why Others Pulled Away

Vietnam and Malaysia clarify what escape looks like.

Samsung’s Vietnamese supplier base grew from 25 enterprises in 2014 to 257 by 2022, a tenfold increase in local supply chain development. Vietnam secured a tariff deal with the United States. Intel committed $7 billion to expand advanced chip packaging in Penang, making it Intel’s largest global production base. Malaysia’s five economic corridors have collectively attracted $162 billion in cumulative investment. The Johor-Singapore Special Economic Zone is becoming a hyperscale data center hub.

What separates these countries from Thailand is policy continuity. Vietnam’s system provides consistent, decades-long investment-friendly policy. Malaysia operates on five-year development plans with dedicated statutory authorities for each economic corridor that function regardless of government changes, even through Malaysia’s own turbulence of three prime ministers between 2018 and 2022. These countries could make long-horizon bets and hold them.

Thailand, with its economic rigidity generating constant political churn, could not. The World Economic Forum found that business executives ranked leadership turnover as the single most problematic factor for doing business in Thailand. The signature digital wallet stimulus was indefinitely postponed in May 2025 with 157 billion baht reallocated after yet another transition in leadership.

Thailand has been middle-income for 37 years. World Bank simulations suggest it will likely remain so past 2050 under current trends. South Korea, starting from a similar GDP per capita in 1960, broke through in under three decades by investing heavily in domestic innovation, R&D, and education. Thailand chose the safe path and has been rewarded with safety and nothing more.

Every time I go back to Bangkok, I notice the same thing: the gap between the people and the system they inhabit. Thais are cosmopolitan, creative, industrious, and increasingly fluent in the language of the global economy.

They are also trapped in a structure that was built by a crisis nearly three decades old, maintained by conglomerates whose dominance predates that crisis, and reinforced by a collective memory that treats caution as the highest virtue.

The standard developing-country story is about building human capital. Thailand already has it. The standard demographic decline story is about rich countries running out of young people.

Thailand is running out of young people without ever having gotten rich. The standard stagnation story is about countries that can’t attract investment. Thailand attracts plenty of investment. It just flows exclusively into what’s already been validated a hundred times over: another mall, another resort, another 7-Eleven.

None of the usual frames quite fit. And that, more than any single statistic, is what makes Thailand’s situation a tragedy rather than a problem. Problems have known solutions. Tragedies are what happen when the ingredients for success are all present and the structure still won’t let them combine.

The Sathorn Unique still stands downtown. Eighty percent complete. Twenty-seven years empty. Nobody lacks the resources to finish it. They lack the conviction that finishing it is safe.

That’s the frame the crisis installed, and it applies to a lot more than one building.

When Sheikh Rashid discovered oil 15 miles offshore in 1966, he had a problem: nobody could see it. So he loaded crude onto a barge, towed it into the Creek, and pumped it into a sand embankment along the shore.

Oil appeared to flow from beneath the city itself. It was pure theater. Dubai never actually had much oil. Abu Dhabi holds 94% of the UAE’s proven reserves. Dubai’s were modest and started declining by the 1990s. Today, oil contributes less than 1% of Dubai’s GDP.

And yet for 30 years, the same Western analysts have kept predicting its collapse. The predictions use economic vocabulary, cite real conditional risks, and arrive at the same conclusion every time: unsustainable.

But strip away the models and what you find isn't economics. It's a story about how wealth is supposed to be earned, who deserves to have it, and what should happen to those who got it wrong.

Dubai keeps violating that story. The predictions keep coming anyway.

The First Thing the Predictions Have to Ignore

To sustain the story that Dubai is a petrostate waiting to implode, you have to skip past sixty-four years of commercial history that precedes the oil.

In 1902, Sheikh Maktoum bin Hasher abolished all customs duties on imports. No tariffs. No processing fees. An open door for anyone who wanted to trade through Dubai’s Creek. When Iran raised its own customs duties a year later, a wave of Iranian merchants relocated from Bandar Lengeh, bringing their families, their trading networks, and their architectural traditions. The coral-stone houses with wind towers that tourists photograph in old Dubai today were built by those Iranian settlers.

By 1949, the pearl industry had collapsed (Kokichi Mikimoto’s cultured pearls had killed the Gulf’s wild-catch market), and Dubai was in desperate shape. People were eating leaves, locusts, and a spiny desert lizard called dhub that tastes like chicken if you’re generous and like sand if you’re honest. The town had maybe 20,000 people and a creek so silted up that large dhows couldn’t enter.

Sheikh Rashid, who took power formally in 1958 but was running things well before, commissioned a British engineering firm to dredge the Creek. Their 1955 feasibility report contains a line of wonderful British understatement: “No records exist of any recent civil engineering works of any magnitude in Dubai and this, coupled with the isolated position of the country, make accurate estimation of costs difficult.” The price tag was £388,000, extraordinary for a town with no oil revenue. He scraped together £200,000 from a local bank and secured a £400,000 loan from Kuwait. When the dredging finished, ships of 500 tonnes could enter. Trade volume jumped immediately.

Edward Henderson, a British oil company representative, described a place where business worth millions was conducted with a wink and a nod, where an expanding town was operating under oil boom conditions fifteen years before any oil was actually discovered. Henderson wrote a line that reads differently in hindsight: “Goodness, what will he be like if he does find oil?”

The free trade zone in 1902. The infrastructure investment in 1959. The oil discovery in 1966. Dubai didn’t need oil to understand commerce. It needed oil to scale what it already knew.

What “Unearned” Actually Means

Why do the collapse predictions keep coming? Because Dubai violates a particular story about how wealth becomes legitimate.

The Western economic narrative is built on a template: labor, sacrifice, deferred gratification, productive struggle. This story has enormous power. It motivates. It organizes. It gives meaning to decades of effort. When a country satisfies the template (Japan’s postwar recovery, Singapore’s ascent from poverty, South Korea’s industrialization), Western analysts call it a miracle and admire it. When a country violates the template, they reach for words like “unsustainable” and “distortion.”

The Gulf violates that narrative. A commodity the world needs sits beneath their ground, and the world pays for it at scale. When wealth appears without the expected suffering, it triggers something that looks like analysis but functions like defense.

People predict collapse. They use economic vocabulary. They cite conditional mechanisms (Dutch Disease, resource curse, rentier effects) that describe real risks under specific conditions. But in ordinary conversation, the conditions drop away and the conclusion hardens: this wealth is unearned, therefore it is temporary.

That smuggles “deserved” into “durable.” It treats moral legitimacy and structural resilience as the same thing, when they are entirely separate questions.

This framing persists even when the data contradicts it. Dubai’s economy runs on wholesale and retail trade (26%), transport and logistics (12%), banking and finance (10%), manufacturing (9%), real estate (7%), construction (6%), and tourism (5%). No single sector dominates. By most measures, it is more diversified than many Western cities of comparable size. And still the analysis defaults to “petrostate.”

A Framework That Broke Inside the Academy

The intellectual scaffolding behind the predictions has a name: rentier state theory. In 1970, Hossein Mahdavy coined the concept to describe pre-revolutionary Iran, a state funded by resource rents from foreign buyers rather than by taxing its own citizens. Beblawi and Luciani formalized it in 1987, arguing that rentier wealth breaks from the “work-reward system” because it comes from “chance or situation” rather than productivity. That sounds like a neutral observation. It is also a value judgment embedded in an economic model. The “work-reward system” is treated as the baseline. Deviation from it becomes pathology.

The critics within the discipline have been saying this for years. Emilie Rutledge, publishing in the OPEC Energy Review in 2017, pointed out that the contemporary notion of rent as “essentially constituting unearned and thus unwarranted income” is divorced from the more accurate idea of ground rent as a legitimate charge for extracting depletable sovereign resources. Scholars treat oil revenue differently from coffee or cocoa revenue, and that difference has nothing to do with economics.

Ozyavuz and Schmid, writing for the French Institute of International Relations in 2015, named the problem directly: “This is where the hint of Orientalism that is sometimes evoked surfaces... analysis of rent in the Middle East appeared to support cultural prejudices.” They called classical rentier state theory “implicitly culturalist, betraying the energy-scarce West’s unease concerning the unequal distribution of energy resources.”

Even Luciani, one of the two scholars who formalized the original theory, eventually distanced himself: “It has never been my understanding that the rentier state paradigm should be either the sole or the overwhelming tool of interpretation.” Among Gulf specialists, by 2019, the original theory appeared “more often as a foil than as a bedrock theoretical perspective.” The framework’s own architects walked away from it.

Outside the academy, in Western media and popular commentary, the theory calcified into a genre. Martin Peretz, then editor-in-chief of The New Republic, wrote in 2008 that Dubai was “a Potemkin village, a house of cards, a mirage” and that it would collapse “soon.” The trigger changed over the decades. The conclusion never did.

The comparison that exposes this most clearly is Singapore. Singapore and Dubai share state-directed economic development, authoritarian governance structures, sovereign wealth funds, zero or low income taxes, a deliberate strategy of becoming a hub, deep dependence on migrant labor, and a founding autocrat whose personal vision drove the whole project.

When Singapore intervenes in markets, it’s called competence. When Gulf states do the same thing, it’s called distortion. Both are state-directed capital. The difference in how they’re discussed tells you more about the person doing the analysis than the country being analyzed.

The 2008 Test

The financial crisis gave the skeptics what looked like vindication.

Dubai had borrowed roughly $80 billion during a four-year construction boom. Property prices had quadrupled between 2002 and 2008. The Atlantis Hotel on Palm Jumeirah opened on November 24, 2008, with a $24 million launch party featuring a Kylie Minogue concert and fireworks visible from space. This was less than two months after Lehman Brothers collapsed.

Subscribe now

By early 2009, property prices in some areas had fallen 50-60%. On November 25, 2009, Dubai World announced it would seek a six-month standstill on $59 billion in debt repayments. European stocks dropped 3%. The Dow fell 155 points. Bank of America warned this could “escalate into a major sovereign default problem, which would then resonate across global emerging markets in the same way that Argentina did in the early 2000s or Russia in the late 1990s.”

For the collapse crowd, this was the ending they had been writing in their heads for years.

Abu Dhabi stepped in with approximately $25 billion in support. The Burj Khalifa opened in January 2010, renamed from “Burj Dubai” at the ceremony to honor Sheikh Khalifa bin Zayed of Abu Dhabi. Everyone understood what the name change meant. Ten months after opening, 825 of its 900 apartments sat empty. Coverage was gleeful. The world’s tallest building as monument to hubris.

Then something happened that the predictions couldn’t accommodate: Dubai recovered.

Dubai World restructured $23.5 billion in debt by May 2010. Nakheel repaid its restructured obligations ahead of schedule and posted AED 860 million in profit by year-end 2010. By October 2012, Burj Khalifa occupancy hit 80%. By 2024-25, it was running between 85 and 90 percent, with capital values reaching AED 3,930 per square foot (a 60% increase since 2021).

Dubai’s population doubled from 1.8 million in 2009 to 4 million by August 2025. Property transactions hit 205,100 residential sales worth $147 billion in 2025. The city attracted more greenfield FDI projects than any other city on earth for the fourth consecutive year.

Consider that the same pattern repeated in miniature with Emirates airline. In 1985, when Gulf Air cut service to Dubai, Sheikh Mohammed started a national airline from scratch: five months to launch, $10 million in seed funding, two planes wet-leased from Pakistan International Airlines. First year: 260,000 passengers and 10,000 tons of freight. Gulf Air’s profits dropped 56%. Analysts called it vanity.

By fiscal year 2024-25, the Emirates Group posted $34.9 billion in revenue and $6.2 billion in profit on a fleet of 260 aircraft. It is the most profitable airline in the world. Jebel Ali Port, dismissed as overkill for a small Gulf emirate, now handles 15.5 million TEUs annually, ahead of Rotterdam and Hong Kong. Capacity built ahead of demand reads as hubris to forecasters who assume demand is fixed. It reads as strategy to anyone who has watched Dubai for more than one cycle.

Mohammed Alabbar, chairman of Emaar Properties, said: “Crises come and go. And cities move on.”

The Story That Doesn’t Update

Sheikh Rashid reportedly told an advisor: “My grandfather rode a camel, my father rode a camel, I drive a Mercedes, my son drives a Land Rover, his son will drive a Land Rover, but his son will ride a camel.” He saw the timeline. The urgency that built Dubai came from that awareness, not from Western op-eds. The response to that urgency was to convert the windfall, relentlessly, into infrastructure, connectivity, and institutional capacity that would outlast the resource.

The real risks exist: concentrated decision-making in a single ruling family, dependence on Abu Dhabi as a financial backstop (as 2008 proved), and a structural reliance on migrant labor with limited political voice. A properly calibrated framework would interrogate those vulnerabilities. Instead, the predictions keep defaulting to “oil will run out” for a city that already ran past oil and kept growing.

The crisis was real. The recovery wasn’t supposed to be. The predictions have been wrong for thirty years because they were never really about economics. They were about sustaining a story of how wealth is supposed to work. And stories, unlike economic models, don’t update when the data comes in.

Subscribe now

49 tweets between 3 and 4 a.m. One every 80 seconds. That was Elon Musk on February 2, 2025, while he was ostensibly leading the federal government’s efficiency initiative.

His friends have begged him to stop. “Get an ice cream cone. Just don’t use Twitter,” pleaded one investor. Grimes, the mother of three of his children, publicly wrote: “I love you but please turn off ur phone.”

His own response: “I think I should not tweet after 3 a.m.”

He kept tweeting after 3 a.m.

I wanted to understand why. Not the pop psychology version, but the structural version.

The Hotel Safe at 3 a.m.

In 2019, Tesla board member Antonio Gracias locked Musk’s phone in a hotel safe. Gracias entered the code himself so Musk couldn’t retrieve it. A well-meaning intervention, the kind of thing friends do when the habit has gotten bad enough that other people can see the damage.

Musk woke at 3 a.m. and called hotel security to crack the safe open. Walter Isaacson, who spent two years shadowing Musk, leads with this anecdote when discussing his relationship to Twitter. Most people hear it and think “addiction.” I did too, at first. Then I looked at the numbers.

In 2018, the Wall Street Journal analyzed 4,925 of his tweets and found it was common for him to post more than 30 times in a single day. By mid-2024, independent analysts measured 67.8 tweets per day over a seven-week span.

Then the 2024 election arrived. Le Monde counted 101 tweets daily between October 5 and November 5. The Washington Post documented 3,870 posts in 26 days (roughly 130 per day), generating over 33 billion views.

During his first 15 days heading DOGE, Irish broadcaster RTÉ counted 1,494 posts, peaking at 178 on a single day. The content shifted as dramatically as the volume: the Washington Post found political tweets rose from 2% of his feed in 2021 to 17% in 2024.

A 2025 academic paper in SAGE Open, analyzing 2,391 tweets through personality modeling, documented the transformation from “neutral and technology-centric discourse” to “politically engaged and emotionally charged presence.”

An independent analyst cross-referenced his posting times with Jack Sweeney’s private jet tracker to estimate time spent on the platform. The conservative number was 3.2 hours a day.

The richest person on earth spends more time posting than most people spend commuting. And the curve is still going up.

The natural response is that he’s wasting his time. That it’s ego, or compulsion, or both. But there’s a problem with that explanation, which is that the returns are enormous and measurable.

And there’s a problem with calling it pure strategy, too: no strategist calls hotel security at 3 a.m. to crack a safe for his phone. The compulsion and the payoff are wired into the same system. Tracing how that system works starts with the money.

The $0 Marketing Department

Tesla, through 2022, spent effectively $0 on traditional advertising. Zero. GM spent $5 billion annually. Ford spent $4.3 billion. Tesla spent nothing. Musk’s personal account (more followers than Ford, Chevrolet, and Chrysler combined) was the marketing department.

The company reached a trillion-dollar market cap on the back of earned media, and Musk’s timeline was the engine producing it. Tesla topped the Wall Street Journal’s list of brands with cultural traction, beating Apple and Nike.

Each individual tweet carried financial weight that entire corporate communications departments couldn’t match. His August 7, 2018, “funding secured” tweet drove Tesla stock up 11% intraday before a halt; expert testimony in the subsequent class-action case estimated $12 billion in aggregate investor harm.

On the upside, a single 2025 tweet detailing Tesla’s AI chip roadmap added approximately $90 billion in market cap in one session. An academic event study analyzing 47 crypto-related Musk tweets found individual posts moved Bitcoin by up to 16.9% in either direction.

No other individual on earth can move this much capital with a sentence typed on a phone. The people closest to him knew it, and they tried to stop it. Tesla investor Joseph Fath compared Musk’s tweeting to Lindsay Lohan’s erratic public behavior and emailed him: “Remember the golden rule. Delete. Don’t Tweet.”

Less than three weeks later, Musk sent the “funding secured” tweet that led to $40 million in SEC fines. The SEC settlement required his resignation as chairman and a consent decree mandating pre-approval of tweets containing material business information, a mechanism people around him called the “Twitter sitter.”

Musk repeatedly violated it. When he tried to terminate the consent decree in 2022, Judge Lewis Liman denied the motion, calling his arguments “meritless.” The Second Circuit upheld the settlement in May 2023. He kept posting.

The hotel safe, the SEC consent decree, the pleas from investors, the judge calling his legal arguments meritless: these are all attempts to build a wall between this man and his phone. Every single one failed. And when you see the returns those tweets generate, you start to understand why.

But advertising replacement and market-moving are the obvious parts. The less obvious part is what happened after he bought the platform.

The Man Who Owns the Acoustics

In February 2023, Musk’s Super Bowl tweet about the Philadelphia Eagles received roughly 9 million impressions. President Joe Biden’s Super Bowl tweet got 29 million. For a man who had just paid $44 billion for the platform, this was apparently intolerable.

Musk flew his private jet to Twitter headquarters that Sunday night. His cousin James Musk sent a Slack message to engineers at 2:36 a.m. flagging a “high urgency” issue. Approximately 80 engineers were pulled in.

By Monday afternoon, the recommendation algorithm included what internal documents called a “power user multiplier” that boosted Musk’s tweets by a factor of 1,000.

When Twitter open-sourced its recommendation algorithm two months later, the code revealed “author_is_elon” as its own category, sitting alongside “Democrat,” “Republican,” and “Power User.”

A literal hardcoded privilege in the system. Internal estimates found over 90% of Musk’s followers now see his tweets in their “For You” feed. Platformer broke the story. Bloomberg confirmed it.

Rather than just owning a megaphone, he owned the acoustics of the room.

The effects were … measurable. Queensland University of Technology researchers found a statistically anomalous engagement boost starting July 13, 2024 (the day of his political endorsement): view counts up 138%, retweets up 238%, likes up 186%, far outpacing comparable political accounts. The AP found his interactions with a German politician boosted her daily audience from 230,000 to 2.2 million.

During the four months around the 2024 election, his posts received 133 billion views: 15 times the audience of any individual elected official, more than 16 times the combined reach of all incoming members of Congress.

But reach is a broadcasting metric. What the algorithm actually gives him is something closer to cognitive infrastructure: 133 billion moments where his framing of an event arrives before anyone else's. When a story breaks on X, Musk's interpretation is often the first one users encounter, and the first frame installed tends to be the one everything else gets measured against. He owns the order in which people receive their thinking.

Political strategists spend millions finding the right word for a policy; “death tax” polls differently than “estate tax.” Musk skips the word choice entirely. He controls the infrastructure through which the words travel, determining which stories get oxygen and which suffocate.

Paddy Leerssen, a researcher at Stanford’s Center for Internet and Society, published a paper in 2025 naming what Musk was doing: “economic instrumentalism,” which he defined as leveraging platform control over public discourse to protect a portfolio heavily dependent on government contracts. Musk’s companies have received $38 billion in cumulative government contracts, loans, subsidies, and tax credits. The posting and the portfolio are connected, and the connection runs through the algorithm.

The Ledger

So there are two sides to this, and the tension between them is what makes Musk so hard to categorize.

On one side: a trillion-dollar company built on $0 in advertising, 220 million followers, 133 billion views in four months, a government appointment, and $38 billion in contracts across his companies.

On the other: a Yale and NBER study (the first rigorous academic analysis of CEO partisan activity’s impact on sales) found his polarizing behavior since October 2022 cost Tesla between 1 million and 1.26 million U.S. vehicle sales. Monthly sales by Q1 2025 would have been approximately 150% higher without the partisan activity. Competitors’ EV sales increased 17 to 22% as customers substituted away. Tesla’s brand value collapsed from $58.3 billion to $27.6 billion, a 53% decline in two years, per Brand Finance.

Then there’s the advertiser exodus. In November 2023, Apple, Disney, IBM, Comcast, Warner Bros., and Paramount all pulled their ad spending from X. Internal documents tracked over 200 ad units from companies pausing or considering pausing. At the DealBook Summit shortly after, Musk told departing advertisers to “go f*ck yourself,” specifically calling out Disney CEO Bob Iger. Then acknowledged the boycott could “kill the company.” U.S. advertising revenue during a five-week span in spring 2023 had already plunged 59% year-over-year.

A million lost car sales and a 53% brand value collapse on one side. A $38 billion government contract portfolio and 133 billion views on the other. Both columns are growing simultaneously, which is exactly what makes the standard explanations fall apart.

Most analysis tries to resolve this tension by picking a side. Either he’s a genius playing 4D chess, or he’s someone who can’t stop posting. I think the more honest read is that both are true simultaneously, and that the inability to separate them is the whole point.

At CNBC, Musk offered his own version: “I’ll say what I want to say, and if the consequence of doing that is losing money, so be it.”

Subscribe now

The Ford Parallel

When people reach for historical comparisons, they usually mention Hearst, Murdoch, or Berlusconi. But the parallel with real structural teeth is Henry Ford.

After losing a Senate race in 1918, Ford bought a newspaper and used it to shape public discourse by owning the outlet directly. The paper reached circulation of 700,000 to 900,000, partly because Ford required every dealership to distribute it.

Every Ford showroom in America became a node in a media distribution network. If you walked in to look at a Model T, you walked out with Ford’s newspaper. A forced-distribution system structurally identical to Musk’s algorithmic self-amplification: you don’t just write the content, you also control the delivery infrastructure that puts it in front of people whether they asked for it or not.

Ford’s “Mr. Ford’s Page” column, where the industrialist spoke directly to a captive audience and bypassed editors entirely, is a direct ancestor of the Musk timeline.

Both men built a manufacturing empire first, then acquired a media channel to protect and extend it. Both understood that owning production wasn’t enough; you had to own the narrative around the production. And both discovered that once you control the distribution infrastructure, there is no external check on what travels through it.

But Ford’s system had limits Musk’s does not. Ford owned factories and a newspaper, but he couldn’t measure in real time which articles resonated, couldn’t adjust distribution speed by the hour, couldn’t suppress a competitor’s message while amplifying his own. He never held a government appointment.

Musk owns the platform, the algorithm, the content, and the audience simultaneously. He can post, boost the post, suppress competitors, and measure the results in real time, all within one system.

Hearst owned newspapers but not factories. Berlusconi held office and broadcast licenses but didn’t manufacture anything. Nobody before Musk has fused media ownership, industrial production, government authority, and personal celebrity into a single integrated operation.

The Engine

Former Tesla employee Christian Marcus described on LinkedIn how this integration actually worked day to day:

“When Elon Musk was informed that we were about to implement a new solution, he would respond in his Twitter feed and let our customers know... This created a powerful communication strategy that made it seem like we were able to react to customer needs and implement changes at lightning speed.”

The company looked impossibly responsive because the CEO’s timeline was synchronized with the engineering team’s deployment schedule. The tweets were coordinated announcements disguised as spontaneity.

That coordination extends across the entire portfolio. Tesla’s vehicle fleet generates physical-world data. X generates human discourse data. xAI’s Grok was designed to leverage X’s live information stream. Together they create a proprietary advantage no competitor can replicate, because no competitor owns both the cars and the conversation.

And the platform keeps getting harder to leave. Government agencies now use X as their primary communication channel, embedding it in public infrastructure. A 2022 paper in Celebrity Studies found that Musk represents “a hyper-celebrified mode of management that has yet to be reproduced by any other executive,” one grounded in “constant conflict” rather than the usual corporate optimism.

A 2025 study in Public Relations Review found that controversy from celebrity CEOs leads to immediate stock declines, but those declines are “partially buffered by high message visibility.”

The more visible you are, the more the market forgives you for being erratic. And Musk is the most visible CEO who has ever lived.

All of it feeds a single loop. More posting drives more followers (his count has more than doubled since acquiring X, reaching 220 million, with no other top account achieving comparable continuous growth). More followers generate outsized influence. That influence converts to political access. Political access protects $38 billion in government contracts across SpaceX, Tesla, and his other companies. And those business resources sustain control of the platform where the posting happens.

As Harvard political scientist Steven Levitsky put it: “This is a combination of economic, media, and political power that I believe has never been seen before in any democracy on Earth.”

Isaacson posed what I think is the right question: “Would a restrained Musk accomplish as much as a Musk unbound?”

The answer is probably “no.”

The same compulsive energy that wakes him at 3 a.m. to break into a hotel safe for his phone is the energy that built a trillion-dollar company on zero advertising dollars.

The same impulsivity that cost a million car sales also produced 133 billion views in four months. The strategic and the pathological are fused, and that fusion is the engine.

The people around Musk have tried everything to separate them. Hotel safes, SEC consent decrees, judicial rulings, public pleas from the mother of his children. None of it worked, and none of it was going to, because the system he’s built doesn’t distinguish between the compulsion and the strategy.

They run on the same circuit, through the same phone, at the same hour of the night.

“I’ll say what I want to say, and if the consequence of doing that is losing money, so be it.”

The phone is where the trillion dollars lives.

In 1989, Alexey Pajitnov had created the most addictive video game on the planet. Tetris was generating millions of dollars across three continents.

His compensation from the Soviet Academy of Sciences: one IBM PC/AT, already obsolete by Western standards.

Pajitnov wasn’t robbed. That framing assumes a thief, a victim, and a shared understanding of what “ownership” means. The category “game designer who owns his creation” didn’t exist. Not suppressed. Not forbidden. Just absent, the way a color you’ve never seen is absent.

The Soviet system had no slot for Pajitnov, and the Western system had no way to explain why.

This is a story about what happens when two worlds collide and discover they aren’t even disagreeing.

The Crowded Desk and the Missing Category

Pajitnov woke between 7:30 and 8 AM, “maybe later, because I worked until midnight every day.” Breakfast was sausage, eggs, and cottage cheese.

He arrived at the Computer Center of the Soviet Academy of Sciences around 10 AM, where his workspace was, in his own words, “extremely crowded, a room built for four or five people seated at desks. On most days it had to accommodate 15 researchers.” He shared his desk with three other people.

The real work happened late, after the desk cleared out.

The Computer Center was busy solving differential equations for outer space and nuclear war. “Nobody gave a shit about this stuff, you know, a small, little game,” Pajitnov said. “Games were something absolutely alien for them.” He built the first Tetris prototype in two weeks on an Elektronika 60 with only 8 kilobytes of RAM, programming in Pascal.

Meanwhile, the KGB was interested in his other work: automatic speech recognition. They wanted to build an audio surveillance system that would start recording automatically when certain “dangerous” keywords were uttered. Pajitnov and his colleagues “tried to avoid, obviously.”

That detail has stayed with me. The KGB showed up at Pajitnov’s crowded desk for the speech recognition work. Nobody showed up for Tetris. The system’s priorities were internally consistent.

The only tangible reward the system ever gave him for creating the most popular video game in history was that single obsolete computer.

He had signed over the rights to the Computer Center for ten years: “It was easier and wiser for me to allow this arrangement than to try to fight for certain personal rights that were almost non-existent during that time.”

And then, remarkably: “You can’t imagine how much internal laws and instruction I overstepped during this time. It’s unbelievable. So if somebody would really look at me, I would be definitely in jail.”

No bitterness. His public stance has been consistently philosophical: “I didn’t make much money at first.” He said he was “happy that so many people were enjoying his creation.”

Pajitnov inhabited a world where the concept of individual ownership of creative software simply did not exist. You can’t feel robbed of something your entire reality has no container for.

A Gray World With No Advertisements

In February 1989, a Hawaiian game designer named Henk Rogers flew to Moscow on a tourist visa, which meant even discussing business was technically illegal. His overwhelming impression was captured in a single word: gray. Gray skies, gray streets, gray and unsmiling people.

The television in his Intourist Hotel room showed only static. The radio played only noise. What disturbed the born marketer most was the complete absence of advertisements: “Nobody was trying to sell you anything!” His brother, who accompanied him on a later trip, “likened it to being in a prison. No colour anywhere.”

Rogers had catastrophic reasons for being there. He’d already convinced Nintendo of America president Minoru Arakawa to bundle Tetris with every Game Boy. Two million dollars’ worth of Game Boy units were already being manufactured, and his in-laws’ property in Hawaii was pledged as collateral.

“I was in deep ‘kimchi,’” Rogers told SpyScape, “because I had 200,000 game cartridges at $10 apiece being manufactured in Japan, and I’d put up all of my in-laws’ property as collateral.” Failure would have destroyed not just his company but his extended family’s financial security.

He couldn’t even find ELORG, the Soviet agency that controlled the Tetris rights. The hotel information desk had no listing. He hired an interpreter from a booth in the hotel lobby who he was certain was KGB but who knew exactly where to go.

Then Rogers did something no Soviet bureaucrat would have anticipated. He simply walked up unannounced and knocked on the front door.

“It was like walking into an ant’s nest because people don’t walk in off the street. Everyone has an appointment and I was supposed to go to a window and present my credentials and I didn’t. I just talked to a random person and said, ‘I’d like to speak to somebody about Tetris,’ and I showed him my little NES version of the game.”

He was promptly interrogated for two hours. Seven people were in the room, “some of them KGB types, being given the third degree.” Rogers’s way of charming police during speeding tickets, he noted, “wasn’t going to work in Moscow. I felt I could end up in a gulag.”

Rogers held up a Nintendo cartridge containing Tetris. To him, it was a commercial product with licensing rights attached to a chain of creators and distributors, and he needed to secure those rights or lose everything.

To the seven officials staring at him, it was something else entirely: unauthorized reproduction of state intellectual property, created by a state employee at a state institution, the rights to which belonged to a state agency.

Same plastic cartridge, two incompatible realities. Thomas Kuhn noticed this same structure in the history of science: where a Newtonian sees a pendulum, an Aristotelian saw constrained free fall.

Same swinging weight, different ontologies. The men in that room weren’t disagreeing about who owned Tetris. They were looking at the same object and seeing different realities.

Foxes, Bodyguards, and a Pidgin for Power

Rogers wasn’t operating inside anyone’s paradigm. He was improvising at a boundary between worlds that had no common language for what was happening. So how did the deal get done?

The answer involves a Soviet bureaucrat named Nikolai Belikov, and it reveals how coordination happens even when mutual understanding doesn’t. Belikov replaced Alexander Alexinko in handling the Tetris negotiations in October 1988, and he was no ordinary apparatchik.

He had built a reputation as “the consummate bureaucratic in-fighter, a master of the art of the well-timed back stab whom you trifled with at your peril. He was the sort of man who was put on the job to do an agency’s dirty work.”

When Rogers walked in unannounced, Belikov didn’t turn him away. He realized “a third party to play against the other two could be a very good thing to have.” Robert Stein, the Hungarian-born middleman who had first brought Tetris to the West, was a “pugnacious” operator whose fatal habit was selling rights before securing them.

He’d sold Tetris PC rights to Mirrorsoft for £3,000 plus royalties before he had any signed agreement from Moscow, maintaining a “Chinese wall” between his Western partners and the Russians, keeping each side ignorant of what the other knew.

Belikov dismantled him by adding a clause to Stein’s contract specifically excluding console rights while distracting him with other changes. “Stein, so focused on just getting the rights, neglected to see the new clause.”

The physicist Peter Galison studied this kind of problem in twentieth-century physics, where theorists, experimentalists, and instrument-makers from entirely distinct communities somehow managed to do productive work together.

His concept was the “trading zone”: a space where two groups can agree on rules of exchange even if they ascribe utterly different significance to the objects being exchanged. Contact languages develop in stages, Galison observed. Jargons become pidgins, pidgins become full-scale creoles.

Belikov was building a trading zone. He didn’t need Rogers, Nintendo, or Stein to share his understanding of what Soviet state property meant. He just needed them to converge on specific terms for specific rights in a specific room. And he found an unlikely counterpart in Nintendo’s Howard Lincoln.

The two were described as “two sly old foxes with far more in common than the gulf of culture, language, and politics that lay between them might suggest.” They “sniffed one another out.”

The signing ceremony on March 22, 1989, had the weight of deliberate theater. Lincoln and Arakawa had traveled to Moscow in secret, telling only two confidants where they were going and using a cover story about visiting Nintendo’s parent office in Japan.

Two heavyweight Soviet bureaucrats attended as witnesses. Nintendo guaranteed ELORG would make at least $5 million. Previous negotiations had dealt in nickels and dimes, “$50,000 here, $100,000 there.” Nintendo’s offer “elevated the discussion to another financial plane entirely.”

But Belikov faced real consequences for building his trading zone. Several influential Soviet officials were displeased. He was questioned by prosecutors. There were rumors of bugs in telephones and men in trench coats trailing people through the streets.

Belikov hired a pair of personal bodyguards. His defense was elegant: Gorbachev said producers needed economic freedom. He was “just doing perestroika.”

Meanwhile, Mirrorsoft’s position had already been torpedoed by Kevin Maxwell, son of media baron Robert Maxwell. When Belikov showed Kevin a Tetris cartridge for the Nintendo Famicom during a meeting, Maxwell had no idea what it was, despite the fact that his own company supposedly held the console rights.

Instead of asserting Mirrorsoft’s claim, he told Belikov he would “entertain a deal” for the console rights, inadvertently confirming that Mirrorsoft’s position was hollow.

When Kevin told his father, Robert Maxwell “went apeshit” and contacted Gorbachev directly. Gorbachev, dealing with a country on the verge of collapse, told Maxwell: “You should no longer worry about the Japanese company.” By then, the contract was signed.

Stein, the man who first brought Tetris to the West, reportedly walked away with only $150,000 to $200,000 total. He “never really got over losing Tetris since he believes it turned friends into enemies.” He died in 2018.

But trading zones don’t always get built successfully. On February 6, 1840, at Waitangi in New Zealand, Captain William Hobson signed a treaty with Māori chiefs. The English text stated that Māori ceded “all the rights and powers of sovereignty.”

But the missionary translating overnight rendered “sovereignty” as “kawanatanga,” a neologism with no prior existence in Māori, while the Māori text guaranteed chiefs “tino rangatiratanga” (full chieftainship).

A bystander known as “Trader Jack” interrupted the proceedings: “Begging your pardon, Sir, but it’s that Mr Williams. He’s not translating a good half of what the Māori say.” In 2014, the Waitangi Tribunal concluded that “the rangatira who signed te Tiriti did not cede their sovereignty.”

No shared object. No pidgin. Just two complete worldviews, each internally coherent, generating a wound that 185 years have not closed.

Belikov’s achievement was building the narrow table at which incommensurable parties could sit. Whether what happened there constituted understanding or merely coordination is a question the participants themselves might not have been able to answer.

Subscribe now

The Pidgin That Worked

Amid all the foxes and bodyguards and collapsing empires, the friendship between Rogers and Pajitnov is the thread that makes this story more than a business case.

When they met during the ELORG negotiations, Pajitnov saw something different from the parade of suited businessmen: “Finally out of all this dressed-in-suit business world, I saw a guy who really liked and understood the game. And somehow we liked each other, almost immediately.”

Rogers recalled: “He was the friend I was looking for in Russia. We got together that night, started talking about game design, immediately jumped into Tetris II. We had stuff to talk about.”

That first night, they drank vodka inside the Pajitnov family’s humble apartment, communicating despite Pajitnov’s broken English and Rogers’s nonexistent Russian. The recognition was professional. “Alexey figured out right away that I was not just some con artist, I was a game designer.”

Rogers noted the decisive difference between himself and Robert Stein: Pajitnov “offered me nothing and asked for nothing.”

Two game designers, from incommensurable worlds, who could not agree on what ownership meant, what a government was for, or what an individual creator was owed. But they could recognize each other across the gap. They could talk about game design. That narrow band of shared understanding became the pidgin through which everything else was eventually negotiated.

For Rogers and Pajitnov, the shared object was the game itself. Trading zones don’t require mutual comprehension. They require something both sides can point to, even if they describe it in different languages. Belikov built his zone out of contract clauses and hard currency. Rogers and Pajitnov built theirs out of falling blocks and vodka.

The long arc confirms it. Pajitnov immigrated to the United States in 1991, settling in Bellevue near Seattle. His work visa was sponsored by Rogers’s company. Six months later, his wife Nina and their sons Peter and Dmitri joined him.

Rogers sold a 15-year license to Tetris to Jamdat (later EA) in 2005 for $137 million, cleverly retaining ownership: “I get the rights back in five years so I can sell them again.” He became one of the wealthiest people in Hawaii.

Belikov, savvy as ever, turned ELORG into a private company after the USSR dissolved and sold his Tetris stake in 2005 for $15 million, with the deal signed in Panama, “in an episode that had much the same spy-movie flavor as the original negotiations.”

The Color You’ve Never Seen

Pajitnov got his computer, then got his freedom, then got his royalties. Rogers risked his family’s home and won a fortune. Stein brought Tetris to the world and died having never recovered from losing it. Belikov became a capitalist. Maxwell lost out.

Kuhn articulated something scholars call “Kuhn Loss”: when one paradigm gives way to another, “problems whose solution was vitally important to the older tradition may temporarily disappear, become obsolete or even unscientific.”

They don’t get answered. They become unthinkable. When the Soviet system dissolved, certain questions (who does creative output belong to collectively? how should a state allocate intellectual work?) didn’t get resolved; they simply evaporated.

At the end of The Structure of Scientific Revolutions, Kuhn asserted that science does not evolve toward anything but only away from something. There is no destination. There is only departure.

The Tetris story works the same way. We didn’t arrive at a resolution where everyone finally agreed on what intellectual property means, who creative output belongs to, or what a game designer deserves. We just departed from one configuration and entered another.

And what stays with me is not who won. It’s the moment before the collision, when each side’s categories were complete and internally coherent and absolutely invisible to the other.

The Soviet system wasn’t evil for failing to compensate Pajitnov. It was something more unsettling than that: it was whole. It had its own logic, its own categories, its own coherent way of making sense of creative work. The slot for “game designer who owns his creation” was absent, the way a color you’ve never seen is absent.

And no amount of arguing, negotiating, or even winning can make you see a color that isn’t in your spectrum.

Only a shared game could do that. And even then, only between two people at a kitchen table, not between two worlds.

On November 17, 2023, Sam Altman clicked a Google Meet link in a Las Vegas hotel room and learned he had five minutes before a blog post would announce his firing as CEO of an $86 billion company.

Microsoft, the company’s largest investor at $13 billion, got sixty seconds of warning.

The board’s entire justification fit into six words: “not consistently candid in his communications.”

What followed was the most unhinged five days in Silicon Valley history: a 52-page secret dossier sent via disappearing emails, a wooden effigy burned in a forest, 97% of employees threatening to quit, and a smoke machine setting off fire alarms at the victory party.

But the real story is what happens when the legal structure of a company and the lived reality of that company become two completely different things. Four people with zero equity tried to blow up a $90 billion enterprise over a weekend.

This crisis exposed a dynamic with implications far beyond tech: structural drift. It’s what happens when an organization’s formal governance framework stays fixed while the lived reality of the organization changes underneath it.

The charter stays the same. The bylaws stay the same. But the organization quietly becomes something else, and nobody updates the paperwork, because everyone implicitly agrees on what the organization actually is.

The formal structure becomes decorative. Then one day, someone tries to use it for its intended purpose, and the whole thing detonates.

That’s exactly what happened at OpenAI. The board tried to exercise the authority the charter gave them. They had the legal right. They had documented evidence. They may have been right on some of the merits. And they discovered that the authority existed only on paper, because the organization described in the charter and the organization experienced by 770 employees, $13 billion in Microsoft capital, and a hundred million users had become two entirely different things.

The question that matters is how the gap between charter and reality grew so wide that nobody noticed until it was too late.

The Structure That Made It Possible

OpenAI was incorporated in December 2015 as a 501(c)(3) tax-exempt nonprofit in Delaware. Its founding donors, including Altman, Elon Musk, Reid Hoffman, and Peter Thiel, pledged $1 billion, though only about $130 million was actually collected.

The charter declared a fiduciary duty not to shareholders but “to humanity.” The mission: ensure artificial general intelligence benefits all of humanity.

By 2019, it became clear that donations couldn’t fund the billions needed to train frontier AI models. So the board created something baroque: a “capped-profit” subsidiary called OpenAI Global LLC, managed by the nonprofit through a shell entity. Investors and employees could earn returns, but those returns were capped at 100x for the first round, with anything beyond the cap flowing back to the nonprofit.

The operating agreement contained a sentence that still makes my jaw drop every time I read it: investors should view their money “in the spirit of a donation, with the understanding that it may be difficult to know what role money will play in a post-AGI world.”

The nonprofit’s six-person board was the supreme governing authority over everything. Microsoft held a roughly 49% profits interest but zero governance rights.

No investor had a board seat, a veto, or any protective provision. The board’s fiduciary duty ran to the charitable mission, not to profit generation.

This structure made perfect sense in 2019. It made progressively less sense every month after ChatGPT launched in November 2022 and reached a hundred million users in two months.

By late 2023, the structural drift had become a chasm. The board governed a nonprofit with a safety mission that happened to run a commercial subsidiary.

The employees, the investors, and Microsoft worked at an $86 billion technology company that happened to have a nonprofit parent. Both versions were internally coherent. Both were defensible on their own terms. They just couldn’t coexist in the same boardroom.

And the board that governed one of the fastest-growing companies in history? It included Adam D’Angelo, the Quora CEO whose AI platform Poe directly competed with OpenAI’s GPT Store, a conflict no one addressed. It included Helen Toner, a Georgetown academic who joined the board through effective altruist networks. It included Tasha McCauley, who co-founded Fellow Robots and taught robotics at Singularity University.

None of them had experience governing a company at this scale.

And it included Ilya Sutskever, OpenAI’s chief scientist. Sutskever had co-invented AlexNet under Geoffrey Hinton at the University of Toronto and catalyzed the deep learning revolution. He was OpenAI’s intellectual core.

He was also the person who, at a September 2022 leadership offsite at Tenaya Lodge in the Sierra Nevada, commissioned a wooden effigy from a local artist. It represented a “good and aligned AI” that turned out to be “lying and deceitful.”

Researchers in bathrobes formed a semicircle beneath ancient redwoods. Sutskever doused the effigy in lighter fluid and set it ablaze. They watched in silence. At the 2022 holiday party, he led employees chanting “Feel the AGI!”, a moment so memorable that staff created a custom Slack reaction emoji. Three employees told reporters he had started “acting like some sort of spiritual leader.”

These four people had the legal authority to fire Sam Altman with zero notice. That authority was about to collide with reality.

A Year of Waiting and a 52-Page Bomb

The coup did not happen overnight. In his sworn October 2025 deposition in the Musk v. Altman lawsuit, Sutskever revealed he had been contemplating Altman’s removal for at least a year before it happened.

The problem was board composition. With Reid Hoffman, Shivon Zilis, and Will Hurd on the board (all considered Altman-friendly), the votes weren’t there. But across 2023, all three departed: Hoffman over conflicts with his own AI investments, Zilis quietly, and Hurd for a quixotic presidential campaign. Their seats were never filled because the remaining directors couldn’t agree on replacements.

“I was waiting for a moment when the board dynamics would allow for Altman to be replaced,” Sutskever testified. “That the majority of the board is not obviously friendly with Sam.” The chief scientist of one of the most consequential organizations in the world spent a year watching board seats empty, waiting for the arithmetic to tip. This was patient.

In October 2023, the catalyst arrived. Sutskever and CTO Mira Murati approached the independent board members with what Toner described as “really serious” conversations.

They used the phrase “psychological abuse.” They said Altman cultivated “a toxic culture of lying.” Murati reportedly told the board: “Altman had a simple playbook: first, say whatever he needed to say to get you to do what he wanted, and second, if that didn’t work, undermine you or destroy your credibility.”

Sutskever added: “I don’t think Sam is the guy who should have the finger on the button for AGI.”

Sutskever compiled his evidence into a 52-page memo. Its opening line: “Sam exhibits a consistent pattern of lying, undermining his execs, and pitting his execs against one another.” He sent it only to the three independent directors using Gmail’s disappearing-message feature.

When asked in his deposition why he used self-destructing emails: “Because I was worried that those memos will somehow leak.”

When asked why he didn’t send it to Altman: “Because I felt that, had he become aware of these discussions, he would just find a way to make them disappear.”

Toner, in a May 2024 TED AI Show interview, laid out the specific instances behind the board’s six-word justification. Two stand out for their concreteness.

When ChatGPT launched in November 2022, the most consequential product launch in AI history (maybe in history, period), the board was not informed in advance: “We learned about ChatGPT on Twitter.”

And when OpenAI’s restrictive agreements threatening to claw back departing employees’ vested equity became public, Altman said he had not been aware of the specific provisions. Vox later obtained incorporation documents from April 2023, bearing his signature, that explicitly authorized them. Whether that reflected genuine unawareness of boilerplate language or something else, it became another data point in the board’s case.

Toner cited additional instances involving the OpenAI Startup Fund and internal safety processes. “We just couldn’t believe things that Sam was telling us,” she said. “And that’s a completely unworkable place to be in as a board.”

Here is the detail that keeps nagging at me. Much of the memo’s content came from a single source: Mira Murati. Sutskever admitted under oath that he never independently verified several key claims, including allegations that Altman had been “pushed out from Y Combinator for similar behaviors.”

When asked why he didn’t verify with Brockman directly: “It didn’t occur to me.” In hindsight, he conceded, “secondhand knowledge is an invitation for further investigation.”

This is where process collapsed.

Whatever the merits of the underlying concerns, a 52-page document built partly on unverified secondhand claims, distributed via disappearing messages, used to justify the termination of a CEO running an $86 billion company with no advance notice to investors, is not a process. It’s a conspiracy. And I don’t mean that pejoratively.

The board members genuinely believed they were protecting humanity. They were operating entirely within the version of OpenAI described in the charter: a nonprofit whose board had unchecked authority and a fiduciary duty to all of mankind.

Within that version, their actions were logical.

The problem was that almost nobody else inhabited that version anymore.

Subscribe now

When De Jure Met De Facto

Thursday, November 16. The four dissenting board members held a furtive video call and voted to fire Altman. That evening, Murati was informed she would become interim CEO.

Friday, November 17, noon. Altman clicked the Google Meet link. Sutskever read a prepared statement. Altman recalled: “It was very strange. I couldn’t get any questions answered. I really did in that first moment think I was having some crazy vivid dream.” Nineteen minutes later, Brockman was told he was removed from the board but “vital to the company.” By 12:30 PM, the blog post was live. Altman’s email and computer access were immediately deactivated.

That evening, as he prepared to fly from Las Vegas back to San Francisco, Oliver Mulherin put his hands on Sam’s shoulders at the foot of the airplane stairs: “Sam, I’m with you 100 percent no matter what, but what the f*ck did you do?” Altman: “Gosh, no!” Oliver: “Okay, I’m good.” They boarded the plane.

Saturday, November 18. Altman had a call with two board members who floated his return. His first reaction: “Absolutely not.” His second: “I will come back if all of you resign right now.” He later called this “not a constructive thing” and said a more diplomatic response might have ended the crisis that morning.

That evening, Altman tweeted “i love the openai [sic] team so much,” and employees began quote-tweeting it with heart emojis en masse, a coordinated signal of loyalty, a headcount of who would follow him out the door. It was most of the company. The first tremor of de facto power making itself felt.

Sunday, November 19. Negotiations, mediated remotely by Satya Nadella, continued all day. The board agreed in principle to resign, then reversed course and installed Emmett Shear as interim CEO.

Shear, the former Twitch co-founder, was on paternity leave with a nine-month-old son. He estimated his personal probability of AI-caused human extinction at between 5% and 50%. He was reportedly a character in a Harry Potter fanfiction popular within the effective altruism community. He accepted after reflecting “for just a few hours.”

Meanwhile, the board had already been rejected by GitHub CEO Nat Friedman and Scale AI CEO Alex Wang. They contacted Anthropic CEO Dario Amodei about replacing Altman and proposed merging the two companies entirely. Toner was “the most supportive” of the Anthropic merger. Sutskever was “very unhappy about it.” Amodei declined both proposals.

Then Nadella made the single most decisive move of the entire crisis. He publicly announced that Altman and Brockman would lead a new Microsoft AI research team, with positions available for every OpenAI employee at matched compensation. Microsoft calculated the cost of absorbing the entire team at approximately $25 billion.

Nadella later called it “in a world of bad choices... definitely not my preferred thing,” but “the worst outcome would have been all these people leave and they go to our competition.”

The announcement gave Altman leverage, gave employees a credible exit, and applied existential pressure on the board. Microsoft stock rose 2% to near all-time highs. The board, which had de jure authority, suddenly had no moves left against the de facto coalition assembling against it.

Monday, November 20. Altman was shaken awake by Oliver, still groggy from a sleeping pill he’d taken at 3 AM. “You need to check your phone.” Sutskever had posted his regret. The emotional turning point had come the day before.

According to the Wall Street Journal, Anna Brockman (Greg’s wife, whose wedding Sutskever had officiated in 2019 at the OpenAI offices, with a robotic hand serving as ring bearer) went to the offices and cried and pleaded with Sutskever to change his mind. The man who had married the couple had tried to destroy the company the groom co-founded. The weight of that broke something.

The structural drift had been invisible for years. One exchange during the crisis negotiations made it impossible to miss.

When OpenAI executives warned that Altman’s absence would destroy the company, and that this contradicted the board’s mission, Toner responded that destruction “could be consistent with the mission.” Even Sutskever testified he disagreed: “At that point in time, the answer was definitely ‘No’ for me.”

That single exchange captures the philosophical chasm at the heart of everything. To Toner, the mission was the load-bearing element. If the company had to be destroyed to protect it, so be it. To the 770 employees who built ChatGPT, the company was the mission.

Both positions were internally consistent. Both were, within their own framing, rational. They were just operating inside different versions of the same organization, and when those versions collided in that room, everyone present could finally see the gap that had been widening for years.

By Monday evening, 745 of approximately 770 employees (97%) had signed a letter threatening mass resignation. The signatories included Murati, COO Brad Lightcap, and, astonishingly, Sutskever himself.

Tuesday, November 21. Shear told the board he would resign, too, if a resolution wasn’t reached. By 10 PM, Altman was back as CEO with a new board chaired by Bret Taylor, alongside Larry Summers and Adam D’Angelo. Three of the four directors who had fired him were gone. At the offices, someone brought a smoke machine to the victory celebration. It triggered the fire alarm and dispatched two fire trucks.

I couldn’t make this up if I tried.

What the Collision Destroyed

The new board hired WilmerHale to investigate. The law firm reviewed 30,000 documents and conducted dozens of interviews. Its March 2024 conclusions threaded a diplomatic needle: the prior board “acted within its broad discretion” to fire Altman, but his “conduct did not mandate removal.”

The firing arose from “a breakdown in the relationship and loss of trust,” not from “concerns regarding product safety or security, the pace of development, OpenAI’s finances, or its statements to investors.”

Larry Summers reportedly told people privately that the investigation found “many instances of Altman saying different things to different people” but to a degree the new board decided “didn’t preclude him from continuing to run the company.”

The full report was never released. The dueling narratives hardened. Toner and McCauley, in a 2024 Economist op-ed, criticized the secrecy.

The aftermath reshaped OpenAI completely. By October 2025, the nonprofit had become the OpenAI Foundation, the for-profit had become a Public Benefit Corporation, and the 100x profit cap was eliminated entirely. Microsoft held a 27% equity stake. Altman, who had held zero equity during the crisis, was reportedly in discussions for a 7% stake. The valuation soared from $86 billion to $500 billion.

No employees actually left for Microsoft. Not one. The threat of 97% of the company walking out had been the ultimate negotiating lever. It worked without a single person following through.

The human cost ran in a different direction. Sutskever departed in May 2024 and founded Safe Superintelligence Inc., which raised $3 billion at a $32 billion valuation largely on the strength of his reputation.

Jan Leike, who co-led the Superalignment team, left for Anthropic, writing that “We finally reached a breaking point” Murati departed in September 2024 and founded Thinking Machines Lab, which reached a $12 billion valuation within months. Nearly half of all AGI safety researchers had left by August 2024.

The Superalignment team, Sutskever’s signature initiative that had been promised 20% of OpenAI’s compute, was dissolved entirely.

The four board members had the legal authority to fire Altman. They had documented evidence of what they genuinely believed was systematic dishonesty. They may have been right on the merits.

But they underestimated something fundamental: that de facto power in an organization doesn’t live in the charter. It lives in the 770 people who show up every day, the $13 billion in capital, and the hundred million users who depend on the product.

The nonprofit board had de jure authority over an $86 billion enterprise. The de facto power belonged to the employees, the investors, and Microsoft. It always had.

The Safety Mechanism That Ate Itself

Exercising the safety mechanism destroyed the safety mechanism.

The board members who cared most about keeping OpenAI accountable to its mission were removed. The Superalignment team was dissolved. The profit cap was eliminated. The nonprofit became a foundation with an equity stake. Every structural guardrail that the original charter had put in place to prevent exactly this outcome was dismantled in the aftermath of the one serious attempt to use them.

Toner and McCauley warned in their Economist op-ed that “self-governance cannot reliably withstand the pressure of profit incentives.” They were right, though the pressure didn’t come from investors or shareholders.

It came from 745 people who chose their CEO over their board, a partner willing to absorb the whole team for $25 billion, and a structure that was only as durable as the shared belief that sustained it.

That’s structural drift in its terminal phase. The charter described one organization. Reality had built another. When the charter tried to assert itself, reality won, and then reality rewrote the charter to make sure it could never happen again.

Governance documents describe an organization. But the organization people actually experience is built from accumulated practice, shared investment, and the interpretive frames that 770 people carry to work every morning. When those two versions diverge far enough, the document becomes decorative. The lived version becomes the real charter.

Every organization carries formal structures designed for a version of itself that may no longer exist.

Most of the time, the gap is invisible.

It becomes visible only when someone leans on the wall and discovers it’s no longer load-bearing.

In August 1939, Albert Einstein sat down on Long Island and wrote a two-page letter. By the time its consequences finished unfolding, the United States had spent $2.2 billion (about $46 billion today), employed 130,000 people, and built a weapon that ended one war while starting an entirely new kind of fear.

Two pages. That’s it.

No detailed budget. No technical specifications. No multi-year roadmap. Just a few hundred words that converted a chaotic pile of uncertain physics into something a President could act on without looking reckless.

I used to assume history turned on speeches, on grand moments of oratory. Then you start noticing how often it actually turns on paperwork.

A war starts because of a telegram. A company pivots because of a memo. A scientific field reorganizes itself because of a short preprint. The big machinery only moves after someone produces a small artifact that makes movement permissible.

If you care about how power actually moves in the real world, Einstein’s letter to Roosevelt is the ultimate case study.

The Mechanics of the Letter

I spent a lot of time with this document. The original sits in the FDR Presidential Library archives, dated August 2, 1939, written from Old Grove Road in Peconic on Long Island. You can read every word. And what strikes me isn’t the physics. It’s the structure.

The letter opens with a move that’s easy to miss: Einstein immediately attributes the underlying work to others. “Some recent work by E. Fermi and L. Szilard,” he writes, “leads me to expect…”

Enrico Fermi was already famous among physicists. Leo Szilard was less famous to the public, but he mattered here more than almost anyone. Szilard was a Hungarian-born physicist who had been thinking about chain reactions for years and who had the political urgency of an immigrant watching Europe (once again) slide into catastrophe.

Einstein was the amplifier. Szilard was the engine.

Then comes something else that’s easy to miss: the letter mixes scientific possibility with operational guidance. Einstein tells Roosevelt that a nuclear chain reaction in uranium may be possible “in the immediate future,” and that this would lead to power and “the construction of bombs.”

He’s careful about uncertainty. He notes that it is “conceivable” that “extremely powerful bombs of a new type” could be built, but also that such bombs might be too heavy for delivery by air.

That last detail is one of my favorites. In the same paragraph where he helps usher in the nuclear age, Einstein still imagines the bomb as something “carried by boat and exploded in a port.” The future arrives wrapped in the constraints of the present.

On the second page, the letter pivots into supply chains. The U.S. has “only very poor ores of uranium in moderate quantities,” he says, and points to Canada, Czechoslovakia, and the Belgian Congo as important sources.

This is one of the tells that the letter was written by people who were thinking concretely. The letter goes beyond “science might do X” to specify exactly where the operation would choke if X turned out to be real.

The Routing Problem

If you want to understand the letter’s force, you have to understand that it was not inevitable that Einstein would be the signer.

The Franklin D. Roosevelt Presidential Library has a short post about the letter that includes this sentence: because Einstein had a prior personal relationship with the Roosevelts and was internationally well-known, a letter was drafted for his signature.

That is blunt, almost to the point of being uncomfortable. Einstein would get read. That was the entire calculation.

This is a thing smart technical people learn late: in large systems, the constraint is often not truth. It is routing. You can have the right idea and still fail if you cannot get the idea to the person who can authorize motion, in a form that survives the trip.

I think about Barry Marshall, the Australian doctor who in the 1980s couldn’t convince the medical establishment that ulcers were caused by bacteria rather than stress.

He had the evidence. He published peer-reviewed papers. Nothing moved. The field had seen the data, but ordinary proof simply couldn’t reorganize an economy of attention and belief that had calcified around stress as the cause.

So Marshall drank a suspension of Helicobacter pylori, gave himself gastritis, and cured it with antibiotics. He needed something dramatic enough to force a recategorization.

Einstein’s letter solved a similar problem through different means. His name was a form of preloaded legitimacy, a credibility escrow account. You could spend it to buy attention from the highest level of government.

And there’s another routing detail that gets smoothed over: the letter was not immediately delivered.

The National Archives notes that Alexander Sachs, a Wall Street economist and Roosevelt adviser, personally delivered it on October 11, 1939, after the outbreak of World War II delayed the initial attempt.

Einstein and Szilard didn’t just “write to the White House.” They found a person who could walk it into the room. The DOE’s Manhattan Project history describes Sachs meeting Roosevelt, initially struggling to persuade him, then succeeding at a second meeting over breakfast.

The Roosevelt Library’s post adds the most quoted line: after learning the contents, Roosevelt told his military adviser, General Edwin “Pa” Watson, “This requires action.”

That phrase is the hinge. The moment a possibility becomes an administrative obligation.

Category Creation

Why did those three words matter so much? Because until Roosevelt said them, the atomic bomb didn’t exist as a category of legitimate government concern.

This is the part that took me the longest to understand. The “problem” with nuclear weapons in 1939 wasn’t that nobody understood fission. Physicists had been publishing about chain reactions. The problem was that no government had installed “atomic bomb” as a real category of risk yet.

Before the category exists, action looks paranoid. After the category exists, action looks responsible.

Einstein’s letter created the category.

Once “atomic bomb” existed as a legitimate object inside the administrative imagination of the U.S. government, it could be staffed, funded, argued about, and escalated.

The letter didn’t contain the Manhattan Project. It contained the seed of a decision tree. It created an official pathway for nuclear work, which then intersected with other inputs (British conclusions, wartime pressures, organizational restructuring) that changed the slope.

The first action was almost comically modest. The uranium committee met on October 21, 1939, recommended acquiring graphite and uranium oxide, and in February 1940 allocated $6,000 to support chain reaction research at Columbia University.

Six thousand dollars is a rounding error in federal budgets, even back then. It is also exactly what early-stage commitment looks like when a system is still deciding whether a new reality is real. You don’t get the blank check. You get the pilot.

But that pilot existed because the category existed. And categories compound.

In June 1940, the uranium committee was subsumed under the National Defense Research Committee, created at the urging of Vannevar Bush. In July 1941, Bush received the British MAUD Committee report estimating a critical mass of uranium-235 around ten kilograms, enough for an enormous explosion, buildable in approximately two years.

By August 1942, the Manhattan Engineer District was officially created. At its peak: 130,000 workers, $2.2 billion spent, over $30 billion in 2023 dollars.

The letter didn’t cause all of that directly. But it made all of that administratively possible.

Writing as a Control Surface

So what was the letter, functionally?

The more I sit with it, the more I think the answer is: a control surface. Writing, in environments where resources are concentrated, reaches into the future and rearranges what becomes possible.

Einstein asks Roosevelt for one thing: a channel. Someone Roosevelt trusts to coordinate across departments, secure uranium ore supply, and speed experimental work. Big projects rarely start with fully funded programs. They start with a person who can carry the problem across the internal borders of a large system.

Einstein also includes a specific claim meant to make the risk feel immediate: Germany has stopped the sale of uranium from mines in Czechoslovakia that it has taken over. He adds a detail about the Kaiser Wilhelm Institute in Berlin and the son of German Under-Secretary of State von Weizsäcker being attached there.

Even if you don’t know how accurate each micro-claim was at the time, you can see what it’s doing structurally. It turns “Germany might” into “Germany is already acting.” A warning becomes a race.

The letter offered Roosevelt a frame where action was prudence and inaction was the real risk. The memos that move resources are rarely the memos that ask for everything. They’re the memos that make the first step unavoidable, and trust that the rest will compound.

When people ask what “narrative power” means, they often expect messaging tricks, clever framing, and persuasion tactics.

Narrative power operates at a deeper level. It is the ability to make a certain course of action feel inevitable, or at least defensible, under uncertainty. It is the ability to install categories that redirect budgets, careers, and national priorities. It operates through small artifacts that cross institutional borders and make movement permissible.

Einstein’s two-page letter is one of the cleanest examples in modern history.

It is short. It is concrete. It is written in the receiver’s language. It comes with preloaded legitimacy. It asks for one thing: a channel. It grounds the physics in supply chains and bottlenecks. And it creates a category where none existed before.

That combination is rare. When it happens, two pages can make $2 billion feel like the responsible next step.

There’s one more detail worth holding. Einstein became a U.S. citizen in 1940, but the Army denied him a security clearance. He was never part of the Manhattan Project, never set foot in Los Alamos.

The system used his legitimacy to justify action while keeping him far from the execution. Once you set a system in motion, the system becomes its own actor. You can start the chain. You cannot steer every link.

I keep asking myself which decisions are sitting in the “looks paranoid” bucket right now that will, five years from now, sit in the “obviously necessary” bucket.

Somewhere, someone is writing the two-page document that will make the difference.

If you’re working on something that needs to cross an institutional border, I’d genuinely like to hear about it. Hit reply.

The question isn’t why Barry Marshall drank a petri dish of bacteria in 1984. The question is why he had to.

He’d already published papers showing Helicobacter pylori in the stomachs of ulcer patients. He’d already demonstrated the association.

The medical establishment looked at his data and shrugged. “Interesting,” they said. “Not decisive.” Meanwhile, surgeons kept cutting vagus nerves, and pharmaceutical companies kept printing money on acid blockers.

Rather than fighting ignorance, Marshall was fighting an economy. Ulcers were less a disease, and more a business model, a career path, and a definition of what “good medicine” looked like.

Being right wouldn’t be enough. To win, he had to make being wrong more expensive than changing one's mind.

The Shape of Ordinary Proof

In early July 1984, Marshall underwent an endoscopy to confirm he was negative for H. pylori, the bacterium he believed caused ulcers.

Three weeks later, he drank a suspension of the bacteria. Five days after that, he felt bloating and fullness, his appetite dropped, and testing confirmed he had colonized his stomach and developed gastritis, the same infection believed to drive most peptic ulcers.

If you tell this as “one brave doctor proved a point,” it becomes medical folklore. If you tell it as an institutional story, the pattern generalizes: the system was structured so that ordinary proof could not land. That is why he had to turn himself into a walking exhibit.

Marshall’s self-experiment was an assault on deniability. Peer-reviewed data fails in paradigm conflict because it leaves room for the incumbent frame to declare the anomaly irrelevant. “Contamination.” “Selection bias.” “Association, not causation.” “Commensal colonization.”

In a mature field, those words are tools for keeping the map intact. But if a healthy investigator demonstrates baseline negativity, then self-inoculates, then develops symptoms and shows colonization and inflammation on follow-up, the easy objections collapse.

In his Nobel lecture, Marshall explicitly positions the self-experiment as a response to a field that wasn’t treating his claim as a credible causal model. But even that didn’t instantly flip the establishment.

The Economy of Ulcers

To see why Marshall’s evidence couldn’t land through normal channels, you have to understand what ulcers were before H. pylori. Not the disease. The economy.

For much of the 20th century, peptic ulcer disease was treated as a chronic acid problem. Sometimes a stress problem. Often, a surgical problem. Vagotomy was a standard procedure. Surgery was serious medicine, and ulcers were a serious surgical domain.

Then the drug era arrived and strengthened the acid paradigm rather than weakening it. Cimetidine hit the market in the mid-1970s and became foundational therapy. It worked. Patients felt better. The mechanism was elegant. The money was extraordinary.

Once a frame has a clean mechanism and a blockbuster treatment, it stops feeling like a hypothesis. It starts feeling like reality itself.

Against that backdrop, Warren and Marshall’s claim that ulcers were tied to a bacterial infection sounded like a category error.

The stomach was supposed to be a hostile, acidic environment where bacteria couldn’t meaningfully persist. The bacteria model implied that a large fraction of ulcer care had been organized around the wrong kind of problem entirely.

They published early signals anyway. In 1983, they reported unidentified curved bacilli on gastric epithelium in active chronic gastritis. In 1984, they reported that such bacteria were present in most patients with active chronic gastritis, duodenal ulcer, or gastric ulcer, suggesting an etiologic role.

Those papers were asking colleagues to reclassify the disease. That’s very expensive. A worldview employs people. It funds departments. It creates “serious” specialties. It produces prestige.

When ulcers are treated as an acid disorder, a certain kind of competence is rewarded: careful titration, elegant pharmacology, sophisticated surgery for refractory cases.

When ulcers are treated as an infection, the center of gravity moves. The old work doesn’t become worthless, but much of it becomes less central, less heroic, and in some cases less necessary.

This is why early rejection tends to be framed as epistemic but driven by incentives and identity. People don’t experience it as “I am defending my status.” They experience it as “this is not how reality works,” because their professional self is built inside that particular description of reality.

Why They Can’t Switch Sides

A paradigm is a massive machine for manufacturing what you might call symbolic capital: the accumulated expectation that listening to certain people will save time, reduce risk, and prevent embarrassment.

It makes the words of skeptics sound half-believed. Without it, even correct claims must be over-proven and over-proven again.

A significant portion of gastroenterology’s symbolic capital had been accumulated under the acid and stress worldview. If the bacterial story wins, the old “good judgment” gets repriced. People who were cautious within the old map become wrong within the new map. That feels like theft, because their status was earned under the prior rules.

This explains a fact that looks paradoxical until you see the structure: even people who privately suspect you are right may still oppose you publicly. Public adoption requires them to defect from the coalition that grants them legitimacy before the new coalition is stable enough to protect them.

A paradigm is a coordination device. It tells people which facts matter, which risks are tolerable, and what counts as a responsible response. Organizations select for plausibility before accuracy, because plausibility allows coordinated action without embarrassment.

When you threaten the paradigm, you’re threatening the coordination mechanism, not merely a set of beliefs.

The Semmelweis Lesson

In the 1840s, at Vienna General Hospital, Ignaz Semmelweis observed that maternal mortality was far higher in the doctors’ clinic than in the midwives’ clinic. He traced the difference to doctors moving from autopsies to deliveries without washing their hands.

When he instituted mandatory handwashing with chlorinated lime solution, mortality dropped from mid-teens percentages to below 2% within months. If you read this through a modern germ-theory lens, the refusal to adopt his method looks insane. If you read it through the institutional lens of the time, it looks grim but coherent.

Semmelweis could show an intervention worked. He could not supply a mechanism the establishment recognized as real. Germ theory wasn’t installed yet. “Cadaverous particles” sounded like mysticism.

Most importantly, his conclusion implicated his colleagues. The new practice didn’t merely say “we can do better.” It said “your ordinary work has been killing mothers.”

Semmelweis was asking colleagues to accept that the defining ritual of their profession was a vector of death. Far beyond a technical correction, that’s a demand for public self-indictment.

Semmelweis is less a story about arrogance and more a story about how paradigms defend their moral order. The old frame preserved innocence. The new frame distributed guilt.

What Marshall Understood

Marshall did something most “right early” people refuse to do. He paid the cost personally to change the cost structure for everyone else.

He didn’t just submit more papers. He constructed a demonstration that moved the argument from “interpret this dataset” to “watch this happen.” That shift matters because it changes what kind of courage is required of the adopter.

His self-experiment was a piece of borrowed cognition: a pre-assembled sequence that others could deploy without having to fight the full battle themselves. Baseline negativity, ingestion, symptoms, colonization, cure. It condensed an abstract causal argument into a narrative that was easy to repeat.

Narrative functions as uncertainty-reduction technology. Frames spread when they provide repeatable justification. Marshall’s demonstration gave the field something they could use.

Semmelweis couldn’t do that. He lived before the mechanism that would let others narrate his results without the stigma of superstition or accusation. He could mandate practice locally. He couldn’t furnish a grammar that let physicians adopt it without feeling accused or irrational.

By February 1994, the NIH Consensus Development Conference recommended antimicrobial therapy to eradicate H. pylori. That’s a full decade after Marshall’s self-experiment. The new model had to become official before it became normal.

Once the consensus statement existed, adoption no longer required personal heresy and became professional compliance. Marshall and Robin Warren received the Nobel Prize in 2005. The Nobel framing treats their work as transforming ulcer disease from chronic recurrence to a condition that can be cured.

More than a scientific conclusion, it became a reallocation of what counts as legitimate medicine in that domain.

The Cost of Being Right Early

Being right early means you’re fighting a disease model and a prestige model at the same time. You’re asking for admissions that can feel morally catastrophic. You’re threatening the coordination mechanism that tells everyone what “responsible” action looks like.

Cynics will cite the obstacle as ignorance or corruption, but it (usually) isn’t. The obstacle is that your audience doesn’t yet have a script that makes adopting your model feel safe. They can’t use your evidence without a grammar that lets them speak it.

If you’re sitting on evidence that your institution isn’t ready to hear, the question isn’t whether you’re right. The question is whether you can make your map easier to adopt than their map is to defend.

Don’t fight for correctness. Fight to make being wrong more expensive than changing their minds.

Most tech CEOs are misunderstood in ordinary ways. People over-attribute success to charisma, or luck, or a single product decision.

Alex Karp is misunderstood in a stranger way: people misclassify what kind of company he runs, so they reach for the wrong interpretive template entirely.

They evaluate him like a consumer-tech founder. He is not one.

They evaluate Palantir like a data company. It is not one.

The result is that Karp gets filed under “eccentric,” “controversial,” or “ideological,” when the more accurate file is “institutional operator selling cognition at scale.”

The Commitment Signal

Read Palantir’s S-1 filing and you’ll find language that would make any normal tech PR team faint. “We have chosen sides,” the company declares, framing its software as a duty to “keep soldiers safe” and explicitly tying its product to “our most vital institutions.”

In a market where CEOs are trained to sound values-neutral and customer-agnostic, Karp narrows the moral and geopolitical frame, polarizes the audience, and keeps going.

If you interpret that as a personality quirk, you miss what it’s actually doing:

This is a commitment signal. It constrains optionality on purpose.

It tells institutional customers (often the state or state-adjacent organizations) that Palantir won’t treat them as an embarrassing phase to grow out of. It tells prospective employees this company won’t rebrand the moment the cultural climate shifts.

In markets where credibility means surviving consequences, Karp is making himself legible to the people who buy consequences.

Two Kinds of Software

To see why this works, you need to understand two kinds of software and why Palantir is playing a different game entirely.

Call it Type A versus Type B.

Type A software is built for clean environments. It wins by being simple, self-serve, and scalable. The customer adapts to the product.

Type B software is built for hostile environments. It wins by being adaptable, secure, and embedded. The product adapts to the customer.

“Hostile” here doesn’t mean violent. It means high stakes, high friction, and high constraint: legacy systems, compliance, coalition politics, adversarial threats, operational urgency.

Palantir is Type B. According to their 2024 annual report (the most recent annual report available at the time of this writing in early 2026), 55% of revenue came from government customers and 45% from commercial.

That alone should push you away from the default Silicon Valley mental model. This is not a company monetizing attention or convenience. Its center of gravity is institutional adoption, procurement, compliance, and long-duration missions.

Once you build for that world, you start to look strange to the rest of tech, because you stop optimizing for what the rest of tech optimizes for.

If you think Palantir is a dashboard company, the pricing looks insane. If you think it’s a data broker, the customer relationships look wrong. If you think it’s an AI lab, the deployment style looks too operational.

Palantir is closer to an operating system for institutional reality, which is not a category most people are used to naming.

This is why Palantir pioneered the “forward deployed engineer” model: sending strong engineers into the field to make software work inside the customer’s messy reality, then iterating based on what breaks.

Karp made this bet early and refused to abandon it, even as the rest of the industry chased product-led growth and self-serve scale.

This is the opposite of the modern SaaS fantasy where products sell themselves and customers are interchangeable. It’s expensive, slow, and deeply sticky. Which is exactly why it fits institutions that treat failure as catastrophic and adoption as political.

In July 2025, the U.S. Army announced an enterprise service agreement allowing the Army and other DoD agencies to purchase Palantir products over up to ten years, with a spending cap of $10 billion.

A cap isn’t a guarantee, but the form matters. This is the Army treating Palantir less like a vendor and more like enterprise infrastructure you want easy to pull off the shelf.

The Legitimacy Engine

Once you accept all this, Karp’s communication style snaps into focus.

Silicon Valley culture treats likability as a proxy for legitimacy. That works when your market is consumers and your distribution is persuasion. In institutional markets, legitimacy is closer to permission.

Permission means the institution can defend the purchase to oversight committees, auditors, and its own sense of identity. The buyer has to be able to explain itself after it buys you.

Karp’s real job, then, looks less like “run a software company” and more like “run a legitimacy engine that lets a certain class of institutions adopt certain capabilities without losing face.”

That is a very different CEO job than “build an app people love.” It requires different language, different recruiting, different tolerance for controversy, and different time horizons.

The irony: the very traits that make Karp hard to market to mainstream tech culture are the traits that increase trust with institutions.

Institutions are allergic to vibes.

They want continuity and commitment.

Karp’s refusal to sand down his worldview is part of how he sells that commitment.

Why It Works

Rather than just the Palantir story, it’s a template for any company (or their founders) trying to sell capabilities, not conveniences, to institutions.

Consumer-tech founders optimize for broad appeal. Institutional operators optimize for trust among buyers who face real consequences. The playbook is different: commitment signals instead of PR polish, embedded deployment instead of self-serve scale, controversy tolerance instead of universal likability.

The companies that win the next decade in institutional markets may look “wrong” through the consumer-tech lens. Too serious. Too opinionated. Too tied to specific customers. Too willing to alienate broad audiences.

But that is not a bug. That’s the strategy.

Most CEOs are misunderstood because people misjudge their skill or luck.

Karp is misunderstood because people misjudge his market.

When Palantir takes heat for controversial contracts, that doesn’t hurt them with the customers who matter. It helps.

It proves to defense buyers and intelligence agencies that this company won’t buckle under activist pressure. It proves the commitment signal is real.

The controversy, the polarization, the refusal to soften his worldview: these aren’t liabilities in his market.

They’re the product.

Sam Altman is unusually good at attracting suspicion.

Depending on who you ask, he’s a visionary trying to steer an unstable technology, or a power-maximizer using “AI safety” as cover. Every news cycle produces a new round of cynical takes: too polished, too political, too eager to sit at the center of the system.

What almost none of those takes do is treat him as a deliberate strategist over a 15-year arc.

Because there is a concrete, traceable story here: how a twenty-something with a decent-but-not-spectacular exit moved laterally, again and again, into roles that controlled distribution, capital, and eventually an entire AI super-cycle.

Whether your feelings are admiration or not, you don’t have to approve of the outcome to learn from the method. And you should.

Altman has, helpfully, left a paper trail. In 2020, in an essay titled The Strength of Being Misunderstood, he argued that the question isn’t “How do I stop caring what people think?” It’s “Whose opinions should I care about and on what timescale?”

Most people, he wrote, figure out the first but neglect the second.

He says the most impressive people he knows care deeply about what others think, but on a very long horizon. They’re willing to let “the newspapers get it wrong” so long as “the history books get it right.”

That frame explains quite a lot about the career that followed.

Why Combinator?

In March 2019, Altman faced a very binary choice.

For a little over three years, he’d been running two organizations at once:

1. Y-Combinator, where he’d become president in 2014, gave him access to over a hundred startups per batch and hundreds per year.

2. OpenAI, which he’d co-founded in 2015, was still mostly a research lab, even as it became more than an academic exercise.

YC’s leadership told him he had to pick one: commit full-time to YC or commit full-time to OpenAI.

He chose OpenAI.

To most people in tech, this looked insane. “President of YC” was the most powerful position in startup land. Direct access to hundreds of companies per year, the ability to shape an entire generation of founders, and a platform that had minted some of the most valuable companies of the decade.

He walked away from it for an odd hybrid: a nonprofit-controlled, “capped-profit” AI lab that still looked more like a research shop than a product company.

Even if it managed to raise serious capital, it was still fundamentally competing with DeepMind, Google Brain, and FAIR on research.

The conventional wisdom was clear: you don’t give up the top of the mountain to bet everything on a weird research lab.

Altman bet on it anyway.

Choosing Platforms Over Exits

The standard path to influence in Silicon Valley is simple enough: get into a rocketship early, ride it to a huge exit, and let the combination of wealth and “I was there” carry you into every important room.

Altman’s path diverged from that script early.

He co-founded Loopt in 2005, went through YC himself, and spent years grinding on a location-based social app that never quite hit escape velocity. In 2011, while still running Loopt, he joined Y Combinator as a part-time partner.

Loopt was sold in an acqui-hire for about $43.4 million in March 2012. With dilution, terms, and the like, his take-home was likely around $4-6 million. Life-changing money, but not “you now own a large fraction of the internet” money.

At that point, the obvious move would have been: do it again. Start another company, knowing more and having more.

He didn’t do that. He used part of the Loopt proceeds to launch Hydrazine Capital, an early-stage fund focused heavily on YC companies. He bought himself a position slightly up the stack, allocating capital across many bets instead of concentrating everything on one.

Then in February 2014, at 28, Paul Graham asked him to become president of YC.

This wasn’t a difficult decision. This was Paul Graham handing him the most powerful platform in startup land. Of course you say yes. Direct access to hundreds of founders per year, the ability to shape how an entire generation thinks about building companies, and a reputation-compounding machine that had already minted Airbnb, Dropbox, and Stripe.

What was unconventional wasn’t taking the role. It was how he used it.

Most people would have treated YC as a job. A prestigious job, but still: show up, run the batches, give advice, post on Twitter, collect the salary and carry. Altman treated it as infrastructure for something else.

You can already see the pattern forming:

• Trade operator role in one company for a platform role across many companies

• Trade clean ownership for messy, leveraged influence

• Trade near-term clarity for long-term optionality

That pattern repeats.

The Multiplier

Altman’s core ethos is simple:

Influence compounds differently than capital.

Capital compounds through returns on financial assets. Influence compounds through network effects.

With money, you can put dollars into deals and hope to get more dollars out. With access and a platform, every interaction is a potential edge: introductions, information flow.

YC was a perfect laboratory for this.

Every batch, Altman met dozens of founders at the most malleable stage of their careers. He saw everything: the raw deck, the dysfunctional founding teams, the early flashes of product-market fit. He had leverage over what they built, who they met, and how they thought about ambition and risk.

A lot of those companies disappeared. But some became Instacart, DoorDash, and Coinbase. Their founders were the ones who remembered who took them seriously when they were pitching half-finished slides to a conference room on Pioneer Way.

Notice the selectivity. Altman didn’t help everyone. He helped people who were ambitious, building something, playing a long game. People who might be important later.

This created a particular kind of reputation. People don’t say, “Sam helped me because he calculated I’d be useful to him later.” They say, “Sam helped me when no one else would.” Both things are true. The calculation doesn’t diminish the value of the help.

From the outside, this looks like “being generous with your time,” as Bay Area figures often are.

From the inside, it’s a very specific kind of leverage:

• Each founder you help is a future allocator of equity, attention, and introductions

• Each company that works becomes a node in your personal distribution network

• Each batch increases both the number of nodes and the density of edges between them

Most people treat “networking” as a set of social obligations. Altman treated YC as a compounding system.

Strategic Opacity

How comfortable Altman is being misread in the short term matters as much as any specific bet he’s made.

In The Strength of Being Misunderstood, Altman argues you should be very selective about whose opinions you weight and on which topics, and that you should care about those opinions on a very long timescale. You should be willing to trade “short-term low status” for “long-term high status,” which means being patient about looking foolish or wrong.

Once you read that, his public behavior looks less enigmatic.

Altman will talk at length about AGI timelines, universal basic income, or the need for more energy. He’ll lay out his views on what a healthy tech ecosystem looks like. But when it comes to concrete maneuvers (why he backed a particular structure, what exactly he promised which investor, what line he’s walking with regulators), the record gets much thinner.

That isn’t an accident. It’s a division of labor:

• Worldview: broadcast widely; invite allies; build consensus over decades

• Tactics: keep quiet; let results, not blog posts, explain the moves

This creates something valuable: interpretive room. When Altman makes a move, different people can read different motives into it.

When the OpenAI board ousted him in November 2023, he didn’t immediately go public with a detailed counter-narrative. He let employees, Microsoft, and other stakeholders make the case.

This allowed multiple versions of the story to coexist, each one serving his interests with different audiences.

His public statements on AI safety follow the same pattern. He expresses enough concern to be taken seriously by safety-focused researchers, but remains optimistic enough to keep building aggressively. Both AI accelerationists and AI safety advocates can find something in his public position to work with.

Compare this to someone like Elon Musk, who over-explains as a rule. Musk will tell you exactly what he thinks about everything, in real-time, often emotionally. This creates clarity but eliminates optionality.

You always know where Musk stands, which means you can predict him, oppose him, or potentially box him in. Altman doesn’t give you that opening.

Lesson: Be open about your worldview and frameworks. Be silent about your tactics and specific calculations. Maximum influence, maximum flexibility.

Public Vision, Private Tactics

A big part of why Altman gets read as “calculating” is that he does something most powerful people avoid: he writes down his worldview.

Starting around 2014, his essays began to cluster around a few consistent themes:

• Intelligence as the master lever on the future (the 2015 “machine intelligence” posts)

• The need for a social backstop like basic income in a highly automated economy (2016 YC basic income project)

• Cheap, abundant energy as a foundation for growth (2015 “Energy”)

• A belief that we can and should drive the cost of many essentials down over time (”Moore’s Law for Everything”)

By the time he co-founded OpenAI in December 2015, he had already spent reputational capital on the argument that machine intelligence would be both enormously valuable and potentially dangerous.

This matters for two reasons.

First, it makes his later moves legible to long-term observers. When he pivots from running YC to running OpenAI, it doesn’t look like a random lurch; it looks like the obvious next move for someone who has been talking about AGI and social risk for years.

Second, it increases the surface area of his influence. Every essay is an invitation to a different set of people (researchers, regulators, founders, policy thinkers) to plug into his frame for what matters. The result is a much thicker web of people whose default assumption is “Sam is basically serious about this.”

Critics argue about how sincerely any of this is held. But the mechanism is straightforward: public worldview, private tactics. The former earns him trust and attention; the latter preserves flexibility.

The OpenAI Coup

The OpenAI pivot in 2019 is a textbook application of the method.

In December 2015, while still president of YC, Altman co-founded OpenAI with Elon Musk and others as a nonprofit lab. He was operationally involved from day one: fundraising, strategy, recruiting. For a little over three years, he effectively ran both YC and OpenAI.

Then, in March 2019, he made the high-variance move.

OpenAI created OpenAI LP, a nonprofit-controlled “capped-profit” entity designed to raise serious capital while nominally preserving its charter. Altman became CEO of this new structure. A few months later, in July 2019, OpenAI announced a $1 billion partnership with Microsoft and committed to using Azure as its exclusive cloud platform.

What Altman saw that others didn’t:

• He saw that AI was going to be genuinely transformative, not in a vague futurist sense but in a three-to-five-year sense.

• He saw that the path to influence in AI wasn’t through gradual academic research but through building something people could touch.

• He saw that being early to this transition would matter more than being right about any particular technical bet.

The transformation of OpenAI from pure research lab to product company (ChatGPT in November 2022, GPT-4 in March 2023, the expanded Microsoft partnership in January 2023) were Altman moves. They were pragmatic, commercial, and slightly norm-violating. They made OpenAI matter in a way that pure research never could.

The November 2023 board fight (brief ouster, rapid return) revealed how strong the network around him had become.

Hundreds of employees threatened to walk. Microsoft made it clear that the partnership was with Altman as much as with the corporate shell. The board surrendered quickly.

The network he’d built was strong enough that the company couldn’t function without him.

The Method, Distilled

If you just want to understand the mechanics, the Altman method looks roughly like this:

Choose leverage over ownership. Repeatedly trade clean founder equity for platform roles that sit on top of many companies (YC) or entire technology layers (OpenAI).

Design for network effects. Put yourself where flows converge (deal flow, talent, research, distribution) rather than where one product lives or dies.

Invest in selective generosity. Spend time and capital on people who are clearly ambitious and long-term oriented. Think of every introduction, every early check, as a possible future edge in the graph.

Broadcast worldview; hide tactics. Use essays and public talks to explain what you’re trying to do in broad strokes. Say very little about how you’re actually going to do it.

Accept being misunderstood on a long horizon. Be willing to look opportunistic, over-optimistic, or even hypocritical in the short run if you believe the long-run bet is sound. Trade “short-term low status” for “long-term high status.”

Compound patiently. Don’t expect the payoff in a year or even five. Altman co-founded OpenAI in 2015 and didn’t go all-in until 2019. The real returns (ChatGPT, GPT-4, the board fight) didn’t show up until the early 2020s.

None of these moves is spectacular on its own. Being a part-time YC partner, starting a small fund, co-founding a research nonprofit, stepping into a strange “capped-profit” structure: each step is legible but not obviously world-historical.

What makes the story interesting is the consistency. Over fifteen years, Altman has repeatedly optimized for position over immediate payout, for platform over product, for history books over newspapers.

Most people in tech, when they have a modest win, try to roll it into a bigger version of the same game. Altman switched games. He chose to play for influence first and money second.

He ended up with both.

Note for newsletter recipients: I changed my newsletter’s name from LinkedUp to Sam’s Newsletter to give me room for broader topical coverage in line with how my positioning is evolving (more on that here).

It’s interesting how quickly things change. Three years ago (can you believe it!), when ChatGPT launched, the consensus was obvious: writers were doomed. AI could already write better than most people, and it was only going to get better. If you made your living producing text, you were in trouble.

Last week, Anthropic posted a job listing for a writer. The salary range: $255,000 to $320,000.

This isn’t Anthropic being quirky. In September, OpenAI posted a content strategist role at $310,000 to $393,000. Same month, PayPal posted a “Head of CEO Content” position with a range cited between $170,000 and $292,000. Just last week, Andreessen Horowitz announced an entire fellowship program for what they called “highly functional people who are also very online.”

I tweeted about the Anthropic JD without thinking much: “Basically all bad/okay writers are near worthless now. Now extremely good writers with great taste trade at a huge premium and own the future of content.”

It got 32,100 views. Which tells you something: people felt this even if they couldn’t articulate it.

Why? Why would Anthropic (the company that makes Claude, probably the best creative writing AI in the world) pay someone $320,000 to write? The obvious answer is they need economics expertise. But The Economist doesn’t pay anywhere near this much. Something else is happening.

Shaping Questions, Not Answering Them

I pulled both of Anthropic’s editorial job postings to see what they actually wanted, and the language is revealing in a way I (mostly) didn’t expect.

The Economics & Policy Lead says:

“Anthropic is at the center of critical conversations about AI’s economic and societal implications.”

Interesting. Continuing:

“[The role will] engage with policymakers, think tanks, academic institutions, and media to advance productive conversations about AI policy and economics.”

Not report on conversations. Advance them.

They want someone with “strong instincts for identifying which policy and economic questions will matter most as AI develops” who can “represent Anthropic’s perspective effectively while engaging constructively with diverse viewpoints.”

Read that first part again. Instead of hiring someone to explain what happened, they’re hiring someone to shape what questions get asked in the first place.

Notice something else: Anthropic posted two editorial roles. The other is for AI and Science, at $200,000 to $255,000. Still substantial, but the policy role pays $65,000 more at the top end. That spread tells you what they think is strategically important.

Algorithmic Negotiation

What these companies are paying for is what I’d call “algorithmic negotiation” at institutional scale. Here’s what I mean by that:

Modern discourse operates on two opposing forces. On one side: the specific knowledge that makes someone worth listening to. On the other side: legibility to the right audiences. Understanding how ideas move through institutions, how frames shift, how to make concepts matter.

Most people can do one or the other. Academics produce rigorous but often unreadable work. PR people produce polished but often empty messaging. The valuable skill is negotiating between these forces without capitulating fully to either.

This requires being embedded in discourse. Not watching it, but living in it. Understanding moves and counter-moves. Knowing which frames are played out and which are fresh. Having what Anthropic’s posting calls “strong instincts for identifying which policy and economic questions will matter most.”

More than simple writing ability, that’s predictive judgment about which conversations determine everything else.

AI can produce clean text explaining any position. But it can’t tell you which position to take, or which question to elevate, or how to shift what the institutions that matter think is important.

What used to be a personal skill is now being systematized.

From Skill to Infrastructure

The a16z fellowship makes this systematization explicit, and it’s where the large strategy becomes clear.

How a16z describes what they’re building: “Forward deployed New Media, where team members embed themselves in portfolio companies for critical periods, and execute a launch together.”

In other words, embed and execute - using the term “forward deployed,” a term popularized by Palantir to describe the concept of sending their engineers into literal warzones to deploy their software for government clients.

The eight-week fellowship has a founding cohort that includes James Reina, Chief of Staff to Mr. Beast. Mr. Beast has 450 million subscribers and routinely gets over 100 million views per video. Reina was inside that operation, understanding how it works.

What is a16z doing here exactly? Extracting operational knowledge about how to build massive audiences, then deploying it strategically.

The speaker lineup includes Marc Andreessen and Katherine Boyle (who both spend lots of time online, specifically on X), Lulu Cheng Meservey (who is famous inside the Bay for flipping the script on PR, guiding clients like Sam Altman and Palmer Luckey out of tricky situations), and Dylan Patel from SemiAnalysis (who turned semiconductor analysis into one of tech’s most influential newsletters).

The fellowship page is explicit about placement: “Potential forward deployment to drive impact at an a16z portco.” They’re embedding them directly into portfolio companies to 10x their influence.

A16z calls this “timeline takeover” - they explicitly offer to “win the internet for a day” as a service.

Their goal, in their own words, is building “the best turnkey media operation in venture” to help founders “win the narrative battle online.” The old agency timeline of three to six months to plan a launch is “archaic,” they write; you now must execute “in a shockingly short period of time.”

What a16z is creating is, ultimately, a pipeline that turns narrative understanding into a competitive advantage for their entire portfolio. They describe themselves as “the F1 Pit Crew of Venture” - systematic infrastructure designed for speed and precision.

Cultural Upstream

This explains something that otherwise seems strange: why do mega-successful entrepreneurs who should be far too busy to post spend hours online?

Elon Musk runs multiple companies. Marc Andreessen runs a major VC firm. Both spend enormous amounts of time posting on X. That looks like distraction until you understand what they’re actually doing:

Building infrastructure to control cultural upstream.

Every industry now has a cultural upstream, and it lives on platforms. In tech, it’s largely Twitter/X. In beauty, it’s TikTok and Instagram. In gaming, Twitch and Discord. The specific platform varies, but the principle is constant.

This creates a recursive loop. The platforms don’t just reflect what’s important. They determine what becomes important. Control the upstream and you control what flows downstream.

From Chaos to System

This shift from individual skill to systematic infrastructure isn’t limited to business. In fact, you see it more clearly in politics.

The 2016 election was the first fully social-media-driven U.S. presidential cycle. The outsider candidate won partly because he understood platform mechanics better than his opponent. But it was still somewhat chaotic; his advantage was partly accidental. Wired wrote at the time about how Facebook and Twitter propelled his campaign in ways that hadn’t been systematically planned.

By 2024, everything had changed. Both campaigns were running sophisticated narrative operations. Politico covered how one campaign joined TikTok despite previously trying to ban it - recognizing the platform had become essential to owning the narrative. The Washington Post documented the rapid-response content operations running in real time. GMF called it “the TikTok era” for presidential campaigns. The professionalization was complete at the national level.

This capability is now cascading to local races. What presidential campaigns had in 2016, city candidates have in 2025 (looking at New York City’s mayoral election earlier this month).

Where Power Flows Now

Big picture, I think we’re watching a major reallocation of power.

For most of history, cultural production was controlled by institutions with distribution: newspapers, publishers, TV networks. The internet disrupted that, but for a while, the new equilibrium wasn’t clear.

Now it’s becoming clear. Cultural production happens on platforms, but control requires systematic infrastructure. And not just using the platforms, but building organizational capacity around them.

That’s why Anthropic is paying $320,000 for someone who can be “at the center of critical conversations.” Why a16z is recruiting from Mr. Beast’s operation and forward-deploying fellows into portfolio companies. Why OpenAI is paying nearly $400,000. Why billionaires who should be too busy are spending hours crafting posts for their growing audiences.

They’re building infrastructure for a new kind of power: the power to shape which questions get asked, which frames become standard, what people think matters.

You might find this uncomfortable. We’re watching influence centralize around systematic infrastructure in a way that feels different from what the early internet promised.

Elite institutions are building machinery to occupy the highest levels of cultural upstream, and they’re doing it by hiring the best people at algorithmic negotiation and embedding them (shameless plug: people like me).

But uncomfortable or not, it’s happening. a16z has the language right there on their site: “embed themselves in portfolio companies” for the purpose of pushing their narrative effectively. Anthropic is explicit that they want to be “at the center of critical conversations.”

The bad writers are finished. The okay writers are finished. The content mills are definitely finished.

The people who can identify which questions will matter before they’re obvious, who can engage with institutions while maintaining strategic clarity, who can negotiate with algorithms while maintaining substantive knowledge?

They’re going to own discourse.

And in a world where narrative increasingly shapes reality (where questions determine policies, regulations, and who gets to build) owning discourse means owning outcomes.

The $300,000 salary isn’t for writing.

It’s for power.

And institutions are paying accordingly.