Now Hidden, Now Open

Reading · Karl Marx and Friedrich Engels, The Communist Manifesto (1848)

“The history of all hitherto existing societies is the history of class struggles. Freeman and slave, patrician and plebeian, lord and serf, guild-master and journeyman, in a word, oppressor and oppressed, stood in constant opposition to one another, carried on an uninterrupted, now hidden, now open fight, a fight that each time ended, either in a revolutionary re-constitution of society at large, or in the common ruin of the contending classes.”

The first section of the Manifesto opens with its most famous sentence, and it is a large claim set down without proof. Every society that has existed has been a contest between classes. The next paragraph names the pairs, from freeman and slave to guild-master and journeyman, and says each fight ended either in a remaking of society or in the ruin of both sides.

The sentence was too large even for its authors. In the English edition of 1888 Engels added a note to say it meant all written history, because in 1847 almost nothing was known of how people lived before records began. Forty years on, he had to narrow the line that made the pamphlet famous.

What I keep from the passage is a smaller phrase: “now hidden, now open.” A conflict between classes does not need a strike or a barricade. Most of the time it is hidden, and it runs through wages and prices and the question of who keeps what a working day produces.

On 5 November the US Bureau of Labor Statistics published its figures for the third quarter of 2026. In the nonfarm business sector, the share of output paid to workers as wages and benefits fell to 53 per cent, the lowest since the series began in 1947. For most of the years after the Second World War it stood between 60 and 66 per cent. Over the past year output per hour rose 2.1 per cent and real pay per hour rose 1.0 per cent. Workers produced more, and the larger part of the gain went to capital and not to pay.

Economists do not agree on what the number means. Some, such as Benjamin Bridgman at the US Bureau of Economic Analysis, argue that part of the fall comes from how output is measured, since more of what firms earn now goes to replace software and machines that wear out fast. Others, such as David Autor and his colleagues, trace it to the rise of a few dominant firms that employ few people for what they sell. The first reading makes the fall smaller than it looks, and the second makes it a question of power, which is closer to what Marx meant.

No one marched against those numbers. There was no strike call and no headline about a war between classes, only a line in a government release. That is the hidden fight. It is fought in pay rounds and contracts, in the choice between raising wages and buying back shares, and in the decision to put a new machine where a worker stood.

But the passage asks more of me than a match between an old sentence and a new number. Marx and Engels said each struggle ended in a new society or in common ruin. A falling labour share has reached neither end. It is a slow shift, and a slow shift lets both sides carry on as if nothing has changed.

If what workers produce keeps rising and their share keeps falling, what would an economy and a state have to change for the gain to reach the people who make it, and who would make that change?

Who Drew Down the World’s Oil

Between February and September the world drew 537 million barrels from its oil stocks. OPEC gives the figure in the feature article of its November report. It equals about five days of world demand, and the size of the draw matters less than who made it.

Governments made most of it. In March the International Energy Agency coordinated a release of 426 million barrels from strategic reserves. By September the reserves of the rich OECD countries had fallen by 218 million barrels, and about two thirds of that oil left the tanks in April and May. The United States supplied 133 million barrels and OECD Asia, mostly Japan, 79 million. Europe supplied 6 million. Europe met most of its share by lowering the stocks it obliges companies to hold, and OPEC notes that this frees oil on paper but need not move a barrel. Outside the OECD, stocks fell by an estimated 294 million barrels, and China drew 156 million of them. Companies in the rich countries drew 24 million. In OECD Asia companies added 20 million barrels to their own tanks while their governments emptied theirs.

So the cost of the war’s lost oil fell on public stockpiles. Strategic reserves exist for a shock like this one, and the release did what it was meant to do. But the shape of the draw shows whose oil a crisis spends first. The firms that trade oil every day held on to theirs. In September OECD commercial stocks stood at 2,811 million barrels, slightly above their average of the last five years. The public’s oil went to market and the companies’ oil stayed in the tank.

The American oil was sold, not lent. OPEC’s October report notes that US federal spending fell in the third quarter partly because of sales from the Strategic Petroleum Reserve, which the national accounts record as lower government consumption. The US government sold oil from its reserve while the war held prices high, and to refill the reserve it will have to buy that oil back.

Oil on the water tells the same story from another side. OPEC counts oil at sea as unchanged since February. Floating storage fell, but tankers on longer routes away from the Gulf held more oil in transit. The world did not find spare oil at sea. It found it in the reserves of governments and in China’s tanks.

The draw also left some countries thinner than others. Countries outside the OECD, China among them, held 60 days of forward cover in February and 46 in September. The OECD fell from 63 days to 60. The countries with the least oil in reserve gave up the most, and they go into the northern winter with the Strait of Hormuz still uncertain. In OPEC’s own words, shipping through the strait still carried a geopolitical risk premium in October.

While the importers’ governments emptied their tanks, the largest exporter began to fill its own. Saudi Arabia reports two numbers to OPEC: what it produces and what it supplies to the market. In July it produced 8.14 million barrels a day and supplied 7.42 million. In August it supplied more than it produced. In October it produced 10.56 million and supplied 9.2 million. OPEC does not say where the difference went. Storage is the likeliest place, and if so, the oil the war took off the market is moving from the reserves of the countries that buy oil to the tanks of the country that sells it.

Every barrel drawn will have to be put back, and here the stocks meet OPEC’s forecast. OPEC expects world demand to grow by 2.6 million barrels a day in 2027. In its October report it named one reason: the rebound would be “further bolstered next year by the need to replenish oil inventories at large.” China’s demand falls slightly this year and grows by almost half a million barrels a day next year. The American caverns, the Japanese tanks and China’s reserves will buy oil from the same producers whose output the war cut. Part of what OPEC calls demand in 2027 will be the world buying back the oil it spent in 2026, at whatever the price is then.

The shock of 2026 will show up in the demand of 2027. The public pays for the release twice: once when the reserve is emptied in a crisis and again when it is filled.

Sources: OPEC, Monthly Oil Market Report, November 2026: Oil Market Highlights (p. iv), Feature Article (p. v), Crude Oil Price Movements (p. 2), Tables 4-1 and 4-2 (p. 32), Table 5-8 with Saudi footnotes (p. 55). OPEC, Monthly Oil Market Report, October 2026: Feature Article (p. v), World Economy (p. 14), Table 5-8 with Saudi footnotes.

From Humanity to the Warfighter: OpenAI

Yesterday, OpenAI announced that it had appointed Dan Driscoll as Senior Counselor for National Security Modernization. Driscoll resigned as US Army Secretary at the end of August and left the post on 2 September. He gave no public reason, but the press reported months of tension with Pete Hegseth, the Secretary of Defence, or of War, as the administration now calls the post. Driscoll is a close friend of Vice President J.D. Vance.

In the opening paragraph of its statement, OpenAI says that Driscoll

“will work with OpenAI teams and leaders across governments, industry, academia, and allied institutions to advance practical modernization initiatives that support the buildout of democratic AI. He will also help internal OpenAI teams understand the operational needs of global national security institutions.”

The list of partners Driscoll will work with covers almost every institution on earth, but the statement names none of them. Driscoll will work across governments and allied institutions. Will OpenAI and Driscoll decide which government or institution counts as an ally, or will Driscoll act as a liaison between OpenAI and the US government?

In the middle, the phrase “practical modernization initiatives” puts three abstractions next to each other. The writer put “practical” first because it gives the reader a feeling of something concrete, but it is vague and the writer knew it. The next two terms are abstract too, and they lean on the vague “practical”.

“Modernization” is corporate jargon and procurement language, and “buildout” could mean telecoms or infrastructure or data centres, but we would not know. They do not say.

“Democratic AI” sounds to me like a democratic ideal OpenAI strives for, but the term does the least work in the statement and says nothing. Sam Altman, OpenAI’s chief executive, gives the meaning away further down the same page. He says the United States has an “extraordinary opportunity to work with other countries to spread the adoption of democratic AI.” Driscoll’s work will be across allies, and the democratic part of it will be for select democracies.

Stripped of all the abstractions, the first paragraph could read: “Dan will work for OpenAI.”

In the second paragraph, OpenAI says Driscoll will advise partners on modernising policy and strategy so that AI can be put to practical use in national security.

Reading the first two paragraphs brings into focus that Dan will help OpenAI sell its AI to defence ministries, armies and intelligence agencies across the world, to whomever it considers an ally. The company never uses these words. It wants to modernise. With Dan’s help.

In “Politics and the English Language”, Orwell writes about prefabricated phrases woven together in an order where the writer never has to form a picture and the reader never forms one either. The perfect hedge.

The vagueness of the statement is not random. I would guess the writer and the legal department chose each word with care. An announcement that OpenAI had hired a former Army secretary to sell its models to armies and to companies that make weapons would raise uncomfortable questions. It would also be true: since 2024 OpenAI has worked with Anduril, a weapons company, on anti-drone systems.

Driscoll says: “I’m joining OpenAI to help the United States and our allies put this technology to work responsibly, for the people who keep us safe.”

“Responsible AI” is a term the Pentagon uses. Since 2020 it has had five principles for AI: responsible, equitable, traceable, reliable and governable. I cannot help but see the paradox of an institution whose job is war and defence calling the improvement of weapons, better killing machines, “responsible.” Responsible, for the Pentagon, means a measure of oversight by people and an accurate AI that will hit the right targets and kill the correct people. Driscoll could have said: “to make faster and more accurate and more deadly weapons.” He chose: “responsibly.”

The statement has one plain sentence, and it is Driscoll’s. AI, he says, “will help them keep pace by cutting through paperwork, making sense of more information, and getting the right answer to the right person faster.” It reads like an office. But getting the right answer to the right person faster is also how a drone’s sighting reaches a gun.

When he ran the Army, Driscoll organised “Operation Jailbreak”, a hackathon to help weapons communicate with each other. At Fort Carson in Colorado, in May, hundreds of engineers came together to help the US do what Ukraine did, to make its weapons talk to each other. Driscoll said the “a-ha moment” came to him in Germany, “because the Ukrainians have become such valiant warriors.” He wanted a simpler, more integrated and more effective system for what the Pentagon calls the warfighter.

The Financial Times reported that Driscoll was impressed by a machine-gun-carrying robot made in three days. The same Dan Driscoll now joins OpenAI.

In 2015, OpenAI wrote the following statement:

“OpenAI is a non-profit artificial intelligence research company. Our goal is to advance digital intelligence in the way that is most likely to benefit humanity as a whole, unconstrained by a need to generate financial return. Since our research is free from financial obligations, we can better focus on a positive human impact.”

When I read OpenAI’s statement about Dan Driscoll joining them, I wondered where the focus on positive human impact went. An impact on humanity as a whole. Now they talk about building clever machines for warfighters in select allied countries.

The road between the two statements has dates. In January 2024, OpenAI deleted “military and warfare” from the uses its policies banned. In June 2025, the Pentagon gave it a contract worth up to $200 million. In February 2026, on the day the Trump administration blacklisted its rival Anthropic for refusing some military uses, OpenAI signed a deal to put its models on the Pentagon’s classified networks. Now it has hired the man who ran the Army.

I must ask: when non-allied nations use and pay for ChatGPT, are they then funding technology that could be used against them or, at best, technology that will not be used to protect them?

India is OpenAI’s second-largest market. It had more than 100 million weekly ChatGPT users by February 2026, but India is not a NATO country. The largest non-allied user of OpenAI pays in data and subscriptions. Unless a user switches it off, conversations on free and personal accounts can be used to train the models the Pentagon now uses. There is no separate military model paid for with military money. A subscriber in Johannesburg, Mumbai or São Paulo helps fund the models that the Pentagon runs on classified networks. It is not only subscribers that fund the models but investors, who put $122 billion into OpenAI in March alone. A US$20 subscriber funds a small part, but subscriber numbers help justify investors’ decisions.

Protection is unequal too. OpenAI’s agreement with the Pentagon bars the use of its models for domestic surveillance of “U.S. persons.” Users in the European Economic Area and Switzerland are served by OpenAI’s Irish company and fall under European law. Everyone else, “most users” in OpenAI’s own words, is served by its American company, and American law allows the surveillance of foreigners abroad. Nothing in the agreement protects an Indian, a South African or a Brazilian.

Access can also be taken away. In June, the US Commerce Department ordered Anthropic to block foreign nationals from its two most advanced models, and both models were offline for 19 days. OpenAI’s own plan for other countries, OpenAI for Countries, asks them to invest in “continued US-led AI leadership.” A country that builds its public services on an American model gives Washington the power to cut them off.

There might not be any neutral suppliers when it comes to AI, but OpenAI started as a promise to humanity.

Prices Rose 3.9% in 13 Months. The Fed Should Raise by Half a Point

A year ago there was no CPI release. The US government shut down in October 2025, and the Bureau of Labor Statistics collected no prices that month. So it cannot publish a 12-month change for most of what Americans buy. I based my chart on the last month before the gap. It measures each price from September 2025 to October 2026, thirteen months, using the index levels BLS published on Tuesday.

Over those thirteen months, US consumer prices rose 3.9 percent. Core prices, without food and energy, rose 2.8 percent. Gasoline rose a third and airline fares 27 percent. Hospital services rose 6.5 percent, food away from home 4 percent and electricity 3.6 percent. Shelter, a third of the basket, rose 3.4 percent. Used cars and car insurance fell.

On the same day, the Treasury sold $42 billion of ten-year notes. Buyers demanded a yield of 5.02 percent, and the notes carry a 5 percent coupon. That rate is now fixed until 2036. The government will pay $2.1 billion a year on this one sale, $21 billion before the notes mature. And it must come back to the market every month and borrow at whatever rate buyers ask.

The Fed held its rate at 3.75 to 4 percent in October. Two members, Lisa Cook and Beth Hammack, voted for a quarter-point rise. The Fed meets again on December 9, and the debate now is how large the rise should be. Traders in fed funds futures give an 88.8 percent chance of a quarter point and 11.2 percent of a half point. I think the traders have the likely outcome right. A half point is hard to sell to people who already pay more for fuel, rent and loans, and to a government that pays 5 percent on its debt.

But the facts point to a half point. The Fed’s target is 2 percent on the PCE price index. That index came near it once, at 2.1 percent in September 2024, the lowest since February 2021. This September it stood at 3.6 percent, with core at 3.1 percent. For more than five years the Fed has promised a return to 2 percent and has not delivered it. A quarter point tells the market the Fed will move slowly again. The bond market has already answered that with a 5 percent coupon, and the government now carries that rate for ten years.

Washington Spares the Servers and Bills Everyone Else

On Tuesday Christopher Waller, a governor of the US Federal Reserve, told a conference in London that the Fed should weigh a half-point rise in December if core prices kept climbing at 0.3 per cent a month. Hours later the US Bureau of Labor Statistics reported that they had, for the third month running. The Fed raised its rate to 3.75 to 4 per cent in September and held it in October, when two of the twelve voters wanted another rise. The committee meets again on 8 and 9 December.

Economists at the Federal Reserve Bank of Minneapolis have measured where this inflation comes from. In August they estimated that tariffs add 0.2 to 0.4 points to core inflation on the Fed’s preferred measure. The AI build-out adds about 0.4 points more, because demand from data centres pushed the price of computers and similar equipment up 12.2 per cent in the year to July. The two causes are about the same size.

Washington treats them very differently. In 2025 American importers paid an effective tariff of 4.5 per cent on goods tied to AI and 12.1 per cent on everything else, because exemptions covered about 69 per cent of AI-related imports. On average, an importer of clothes, cars or machine parts paid nearly three times the rate that an importer of AI hardware paid.

A rise in December puts a second bill on the same businesses. The Fed cannot aim its rate at one sector, and a higher rate reaches every borrower. The businesses that meet it hardest borrow from banks and sell to households that borrow too: homebuilders, car dealers, retailers and mid-sized manufacturers. They have paid the higher tariff for a year. Now they will carry most of the cost of fighting an inflation that the lightly taxed AI build-out helped cause, while the build-out goes on.

This is an inference from the Minneapolis figures and the December calendar, not a finding of the Fed. The Fed does not set tariffs, and it raises rates for the whole economy because it has no other tool. The split it produces is a choice made elsewhere, in the tariff schedule.

That schedule now faces a new Congress and voters who have turned against the build-out. On 3 November Americans elected a Democratic Congress, and voters in Texas, Georgia and Ohio backed candidates who promised to rein in data centres. In Ohio, 15 towns and cities passed bans or limits. The new Congress takes its seats in January. The president, for his part, is weighing broader tariffs on semiconductors and the products that contain them, which would close the gap from the other side and raise the cost of the build-out itself.

Either way the gap will not hold. Few voters would accept a schedule that taxes the old economy’s imports at nearly three times the rate of the new one’s, followed by a rate rise that lands on the old economy first, if they saw it set out with the numbers. The December vote decides how much more the businesses outside the boom pay before Congress or the president decides who pays the tariff.

The Vote Against the Data Center

On Tuesday American voters elected a Democratic Congress and in dozens of towns and counties they voted to ban or restrict the data centers that artificial intelligence needs. By Thursday the vote had reached markets around the world. In New York the S&P 500 fell 4.4 percent and the Nasdaq 100 fell 6.3 percent, the worst day for both since the tariff shock of April 2025. On Friday Japan’s Nikkei closed more than 20 percent below its June peak and entered a bear market. The Korea Exchange halted trading for twenty minutes after the Kospi fell almost 9 percent at the open. SK Hynix, which makes the memory chips that go into AI servers, lost a tenth of its value in a day and TSMC in Taiwan fell almost 5 percent.

Voters in ten states decided 41 local measures on data center policy and the builders won only one of them, in North Dakota. Ohio alone held votes in 21 municipalities. Some measures banned data centers outright and others paused building, capped how much power a site could use or required a public vote before construction. The vote for Congress carried the same question. In the NBC News exit poll 56 percent of voters said AI and data centers were a crucial or major factor in how they voted for Congress.

The pattern of the losses shows what investors fear. Seoul, Tokyo and Taipei sold hardest because they make the chips and machines for data centers that may now not be built in Texas and Ohio, or not built as fast. Europe has few such companies and its technology shares barely moved. ASML, the Dutch maker of chipmaking machines, rose 4 percent. In Hong Kong the Hang Seng Tech index rose more than 5 percent on Thursday as Chinese tech companies rallied. Money is not leaving the world economy. It is leaving one bet on it.

The American numbers point the same way. In the month before the election the Nasdaq 100 fell almost 10 percent while the average stock in the S&P 500 rose more than 4 percent. Citadel Securities found the gap has been wider only in the dot-com years. Investors were moving from the giants into the rest of the market. The US economy is not weakening. American employers added 165,000 jobs in October, more than twice what forecasters expected. The US unemployment rate fell to 4.1 percent. The trouble is prices. US consumer prices rose 3.7 percent in the year to September, pushed up by the conflict in the Middle East. Traders now bet the Federal Reserve will raise rates in December. A rise would lift the dollar further and tighten money for every country and company that borrows in it.

So far the evidence points to the fall of one industry and not to a global downturn. The election made that fall sharper because it put a political limit on building. Two events next week will test that reading. On Tuesday the US publishes its October inflation figure, the last the Fed will see before it votes. And Anthropic, the American company that makes Claude, must decide whether to go ahead with a share sale meant to value it at about $1.5 trillion. Three investors told the Wall Street Journal the price must come down, one of them toward $1 trillion. If Anthropic cannot sell its shares at the price it wanted, the money for every new data center costs more, from Texas to Seoul.

For now the world’s investors are not running from risk. They are asking what a data center is worth when the people who live next to it can vote it down, on a town ballot in Ohio and in the vote for Congress.

Why does the Federal Reserve watch how many Americans quit their jobs?

On 3 November the US Bureau of Labor Statistics reported that 3.4 million Americans quit their jobs in September, 321,000 more than in August. The quits rate rose to 2.1% of workers, its highest since June 2025. Hotels and restaurants made up 197,000 of the rise, and quits there jumped from 3.6% to 5.0% of staff in a single month. Employers held 7.5 million jobs open on the last day of September. The Bureau treats the quits rate as a gauge of whether workers are willing and able to leave. The Federal Reserve reads the same number with a different question in mind. It wants to know whether workers can push up their pay.

In the United States the quit is the main way a worker bargains. Only 5.9% of private-sector workers belonged to a union in 2025, so most have no contract and no one to sit across the table for them. A worker who wants more money finds another job and leaves, or tells her boss she will. The threat works when jobs are easy to find. In September there were 7.5 million openings and 7.1 million unemployed people, a little more than one job for each person looking. Every quit leaves a firm with a post to fill, and the firm must pay to fill it or pay to keep the people who stay.

Workers have a reason to look. Consumer prices rose 3.7% in the year to September, up from 3.4% in August. Gasoline rose 9.1% in September alone and made up more than half of that month’s rise. Average hourly pay rose 3.1% in the year to October, to $38.00. Prices have outrun pay, and fuel accounts for most of the gap. Leave out food and energy and prices rose 2.6%, slower than wages. A cook who leaves one kitchen for a better rate in the next street is trying to win back what she lost at the pump.

The Fed has already moved. On 16 September it raised its target rate by a quarter point to 3.75–4%. On 28 October it held the rate, but two of the twelve voters wanted another rise: Governor Lisa Cook and Beth Hammack, president of the Cleveland Fed. The statement said that inflation “remains elevated” and closed with a promise: “The Committee will deliver price stability.” A higher rate does not lower the price of oil. It works through credit. Firms that borrow to grow slow their hiring and post fewer jobs, and with fewer openings a worker who quits has fewer places to go.

Kevin Warsh, who took the chair in May, argued in the Wall Street Journal last November that the Fed should give up the belief that inflation comes when the economy grows fast and “workers get paid too much.” Government spending and money printing cause inflation, he wrote. The Fed’s October statement says nothing about wages and blames no one. But the tool it has used reaches the labour market and leaves the gasoline price where it is.

The case for a wage push rests on thin evidence. One month of quits, most of them in kitchens and hotels where staff turn over fast, does not make a trend. Layoffs fell to 1.6 million in September. In October employers added 165,000 jobs and the unemployment rate held at 4.1%, but the number of people working part time because their hours were cut or they could not find full-time work rose by 339,000 to 4.8 million. The figures point both ways.

The OpenQ

The Bureau publishes October’s quits on 1 December, eight days before the Fed’s next decision on 9 December. If quits keep rising, workers may be starting to win back what fuel took from them. If the Fed raises rates again first, they will have fewer jobs to move to. The open question for the coming months is whether American workers can use the quit to close the gap between their pay and their prices, or whether the Fed raises rates in December and narrows the market in which they bargain.

Think About

The Bureau counts a quit as a sign of a worker’s freedom to leave a job. The Fed counts the same quit as a sign that the labour market is running hot. Nobody who quit a kitchen in September chose the war that raised the price of gasoline. When prices rise faster than pay, someone carries the loss. Should a central bank treat a worker’s power to walk out as a risk to manage?

Ask or Reply

Send your answer or email me a question for a future OpenQ at info@wernermouton.com.

Meta sold four-fifths of its largest data center in October 2025 and agreed to rent it back. The campus is called Hyperion and it is going up on more than two thousand acres of farmland in Richland Parish, in the north-east corner of Louisiana. Funds run by Blue Owl Capital took the stake. A vehicle on the Blue Owl side called Beignet Investor then sold about twenty-seven billion dollars of bonds to pay for the buildings and the power and cooling inside them. S&P rated the bonds one notch below Meta itself. They pay a little more than six and a half percent a year and they fall due in May 2049.

Stijn Van Nieuwerburgh, a professor of finance at Columbia Business School, sets that deal inside a much larger one in a paper for the Brookings Papers on Economic Activity this autumn. He counted the data center projects planned across the United States and estimated how many will be built. On his central case the country will spend about ten trillion dollars on data centers and their chips and power between 2025 and 2032. That is more than three and a half cents of every dollar the American economy produces, in each of those eight years. The railroads of the Gilded Age took a little over two cents. The interstate highways and the fibre boom of the late 1990s took about one cent each. As a share of the economy, the AI buildout is larger than any of the great American building booms he measured.

The companies that want the computing cannot pay for it from what they earn. By the paper’s count, Oracle, Microsoft, Amazon, Meta and Alphabet are expected this year to spend more on buildings and machines than their businesses bring in as cash, for the first time. So the money comes from outside, through leases and joint ventures and private credit and bonds sold by special vehicles like Beignet. Van Nieuwerburgh states the result in one line: “These arrangements transform rather than eliminate risk.”

Hyperion shows how the transformation works, and Meta’s own quarterly filing for September 2026 sets out the terms. The lease begins in 2029. Its first term is four years, and Meta may renew it in steps up to twenty. If Meta walks away, it guarantees the owners part of the value of the campus, up to a cap. The filing then says: “RVG payments are not probable, and therefore no liability has been recorded to date.” RVG means the residual value guarantee. Meta also keeps the whole venture off its balance sheet. Its own analysis found that the choices that matter most to the venture are about finding new tenants and selling the buildings, and that those choices do not belong to Meta.

The terms set two clocks side by side. The bonds run for almost twenty-four years. The lease can end after four. The chips in the halls, on the paper’s assumption, are worth running for about six years. So the party with the shortest commitment decides what the campus is used for, and the long years belong to the people who lent. Meta pays for this. The paper estimates the bonds cost at least a full percentage point more than Meta’s own borrowing, and the lease passes that cost back to Meta. The price buys a balance sheet that looks light, which is worth a great deal to a company valued for its growth.

The long years end up in many hands. Pimco anchored the Beignet bonds, and Pimco invests the savings of pension funds and households. Moody’s counted nearly a trillion dollars of data center lease commitments at the five big cloud companies at the end of 2025, and found that about two-thirds were for leases not yet begun and so not yet on their balance sheets. That sum was larger than all their adjusted debt. In Louisiana the Public Service Commission voted in August 2025 to let Entergy build three gas plants for the campus. The power contract with Meta runs fifteen years and a gas plant runs for decades. And in August 2026 Nvidia signed preliminary agreements with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to raise more than five hundred billion dollars for AI infrastructure. Jensen Huang said Nvidia may guarantee up to a quarter of the value of the equipment in a deal. The company that sells the chips now promises what they will be worth.

The paper also measures what the buildout must earn. For the capacity finished by 2032 to return ten percent a year, it needs revenue of about three point seven trillion dollars a year. By the paper’s estimate that is about thirty-seven times what OpenAI and Anthropic, the two leading model companies, take in together today. The price it needs for an hour of chip time sits near what the market charges now. The risk is whether that price holds once the country has added all this capacity at once.

The strongest case against alarm comes from the same pages and from Meta’s accounts. Meta brought in close to a hundred billion dollars of operating cash in the first nine months of 2026, many times what it owes each year on its debts. The guarantee is real, so if Hyperion loses value Meta pays much of the loss. The bondholders are large professional investors who chose long, steady income at an A-plus rating. Van Nieuwerburgh writes that distress is not imminent. And the old booms left their works behind them. Jay Cooke and Company failed in September 1873 over the bonds of the Northern Pacific Railway and set off a national panic, but the line was finished ten years later and freight still runs on it. Global Crossing filed for bankruptcy in January 2002, and the cable it laid under the oceans went on to carry the internet.

That case is right about the works, and it shows where the losses went. In 1873 and in 2002 the country kept the rails and the cable, and the people who had lent against them took the loss. The AI buildout is arranged the same way with one change. The companies that decide how much to build, and for what, have written contracts that let them leave early. Most of the long years sit with the lenders, the savers behind them, the ratepayers of north Louisiana whose gas plants will outlast the power contract, and now the chipmaker that guarantees its own goods. The paper asks for better measurement and disclosure while the industry’s finances are still taking shape. At Hyperion they have already taken shape.

Meta’s lease on Hyperion begins in 2029 and its first term ends in 2033. The bonds are due in May 2049.

What the Poverty Chart Says About Neoliberalism

In 1950 six in every ten people on earth lived in extreme poverty. In 2025 one in ten did. The World Bank published these figures in October, in a short post by Christoph Lakner and Samuel Kofi Tetteh Baah, and it drew the line at $3.00 a day. Over the same years the world’s population more than tripled, from 2.5 billion to 8.3 billion, and still the number of extreme poor fell, from about 1.5 billion to just over 800 million.

Most of those years belong to the neoliberal era. The German economist Alexander Rüstow coined the word at a meeting of liberal thinkers in Paris in 1938, and he meant a liberalism in which a strong state set the rules of the market. The word changed sides later. Critics tied it to the market reforms of Augusto Pinochet’s Chile, and it became an insult. By the 1980s the program had its leaders in London and Washington. Ronald Reagan fired 11,345 striking air traffic controllers in August 1981. Margaret Thatcher’s government sold British Telecom in 1984 and British Gas in 1986, and it outlasted a miners’ strike that ran for 51 weeks until March 1985. The IMF and the World Bank made adjustment loans to indebted governments on condition that they reformed their budgets and opened their markets. Defenders of the program point to the poverty chart and say the market did this.

The Bank’s own chart tells a narrower story. China cut the number of its people in extreme poverty from about a billion in 1990 to almost none by 2015, and India began its fall about a decade later. Take those two countries out and the global decline looks modest. Neither country followed the Washington script. The Chinese Communist Party opened to trade and foreign investment, but the state kept the big banks and kept the capital account closed. As late as 2005 the IMF urged Beijing to take the government out of bank management. India opened after its 1991 balance-of-payments crisis, with IMF support, but it opened in stages.

The countries that took the full program did worse. Across Latin America the debt crisis of the 1980s brought adjustment programs and what ECLAC called a lost decade of falling income per head. In the first half of the 1980s UNICEF found that real government health spending per person fell in most of Africa and Latin America, and that public spending per person on primary schooling fell in two-thirds of the developing countries with data. Sub-Saharan Africa took more of the program than any other region. Of the twenty countries that took the most IMF and World Bank adjustment loans between 1980 and 1999, William Easterly counted twelve in the region, with 14 to 26 loans each, against an average of seven for developing countries. The reforms set out to phase out subsidies on fertilizer and other farm inputs. Income per head in the twenty grew by about 0.1 percent a year, and the number of people in extreme poverty in Sub-Saharan Africa kept rising until, by 2015, the region held more than half the world’s extreme poor.

Russia took the program fast. It freed prices in January 1992 and transferred more than 15,000 firms through vouchers between 1992 and 1994, and managers and workers ended up with about two-thirds of the shares. Output fell by 45 percent or more, by a World Bank evaluation. Life expectancy for Russian men fell from 63.8 years in 1990 to 57.7 in 1994.

The World Bank published its flagship Poverty, Prosperity, and Planet Report 2026 on the same day as the blog post, and the report measures the stall. In 2026, 811 million people live on less than $3.00 a day, and the rate has not moved since 2019. More than half of them live in nine economies. By 2030 the Bank expects 87 percent of the extreme poor to live in Sub-Saharan Africa, in states it classes as fragile or in conflict, or in both, and without faster growth it projects more extreme poor in 2030 than there are now.

The $3.00 line also hides how little the escape bought. Nearly half of humanity, 3.7 billion people, lives on less than $8.30 a day, and that number has not fallen since 1990. The Bank says most of them are one shock away from falling back. Its Prosperity Gap, a measure of how far the world stands from a secure daily standard of living, halved between 1990 and 2013 and has narrowed by barely 1 percent a year since. Hundreds of millions of people left extreme poverty and stopped a few dollars above it.

The Bank no longer talks about budgets and open capital. It talks about jobs. Nearly every poor household owns one productive asset, the skill and labor of its members, and the report finds that raising poor workers’ schooling to the level of those above the line would lift their earnings by about 40 percent and cut extreme poverty nearly in half. Putting three in four working-age adults in poor households into work, mostly by bringing women into jobs, would cut it by about a fifth, and so would moving poor workers off the farm and into wage work. To bring extreme poverty down to 3 percent by 2050, income per person would have to grow about 4 percent a year, twice the pace of the last 25 years.

In the 1980s the same institution lent to poor governments on condition that they cut public spending, and UNICEF found that spending on schools fell with it. Steger and Roy write that the people who saw the Soviet collapse as the end of history did not see the populists coming, with their promise of a return to the autonomous nation state. The poverty numbers help explain the surprise. The great escape from poverty happened in countries whose governments kept their grip on money and trade, and in 2022 about two in three of the world’s extreme poor lived in Sub-Saharan Africa, the region that took the most adjustment loans.