The New York Times Magazine has a very long article that asks the question, being posed in many quarters, as to whether we are in an AI bubble. The argument that we are hinges on the massive amounts of money that is being spent on data centers and computers estimated by Cory Doctorow to be currently running at about $1.4 trillion while the AI companies are getting revenues of only about $50 billion from subscriptions to their services. So where is the rest of the money coming from? From borrowing, as rich investors pour money into these AI companies in the belief that they will grow rapidly and be profitable at some point. This belief drives up the stock prices. As Doctorow says:

I’m pretty baffled that there’s people who say it’s not a bubble. Let’s just run the numbers again. Trillion dollars in expenditure this year, over $1.4 trillion total, $50 billion revenue. Seven companies, the magnificent seven that are 35% of the S&P 500. Six of them lose money on AI, and the seventh one makes money from the other six. [The ‘other six’ are Alphabet, Amazon, Meta, Microsoft, Oracle, and Anthropic and the seventh is the chip maker Nvidia – MS]
…

That’s Nvidia, yes. They’re all passing around the same hundred-billion-dollar IOU really fast and pretending it’s in all of their bank accounts at once. The assets depreciate really quickly. They amortize those data centers and the GPUs in them over five years. They actually replace them every two to three years. The replacements, pretty intense.
…

GPUs, data centers, and the AI that runs on them, right? They’re saying that these things are going to last them about five years. The reason that matters is if you spend $1 trillion on stuff that’s only going to last five years, then in five years, you’re going to need another $1 trillion if you want to stay in business at least. Nvidia is actually doing some pretty amazing things with new generations of chips every year.

One of the ways they’re getting such performance out of these chips every year is they’re throwing away one of the principles that normally constrains new technology manufacturers, which is backwards compatibility. These new chips, they have really different heat dissipation, networking, power consumption, and often you can’t put them in the same building. You need to basically tear the data center down to the foundation’s lab and build another one for the next generation of chips.
…

These are not durable assets in this model. A third of the stock market, their assets depreciating every three years. They’re spending $1 trillion a year. They’re making $50 billion a year.

But how long can this state of affairs last? That is the question and the New York Times Magazine article tries to address the question by looking carefully at Oracle founder Larry Ellison who entered the field later and seems to have bet his immense fortune as well as borrowing heavily in an attempt to catch up with the other AI companies, who in turn are also spending heavily out of fear of being left behind in the Next Big Thing and so we have runaway growth in spending.

But his big bet on A.I. was built on an astronomical amount of debt in every imaginable form — bonds, letters of credit, asset-backed securities — available in seemingly unlimited quantities, because the more you spent building A.I. infrastructure, the more you would earn, or so the logic went. Computer theorists called it the scaling hypothesis. It held that advancements in A.I. were directly tied to the generation of unprecedented amounts of computing power to process unprecedented volumes of data. Reaching the holy grail of artificial general intelligence, or A.G.I., when computers match or surpass human thinking at any task, was going to require bigger data centers and a lot more of them. It all came down to capital expenditures — capex, in the lingo of the Valley. Whoever controlled the most computing power would control the A.I. economy.

But lately, some investors and analysts have started questioning the scaling hypothesis or at least asking if all this spending is sustainable. The market has been gyrating wildly in recent weeks, as concerns have grown about whether the trillions of dollars being furiously pumped into this global ecosystem of data centers will ever return the promised profits.
…

Ellison is personally worth about $55 billion less than he was on the morning he flew to Washington, and more than $200 billion less than he was at his peak in September. Oracle has pushed the limits of the credit market and is facing steeper interest rates from lenders, and its credit rating has been downgraded to a notch above “junk” status.

Ellison and his hyperscaler peers are confident that all of their borrowing and spending will set them up to dominate a transformed global economy. As a percentage of the nation’s G.D.P., the great A.I. infrastructure build-out is on track to exceed the construction of the American railroad system during the second half of the 19th century, the building of the Interstate highway system 100 years later and the Apollo space program.

The hyperscalers — Alphabet, Amazon, Meta, Microsoft, Oracle — are some of the richest companies in the world, and the stock market is heavily dependent on them for its growth. If Oracle were to falter, the repercussions could be wide-ranging. Americans are more invested in the stock market than ever before, and the A.I. boom has been driving a disproportionate amount of America’s economic growth. Last fall, Gita Gopinath, a former chief economist at the International Monetary Fund, writing in The Economist, estimated that an A.I. crash would wipe out $20 trillion in American wealth — far more than the dot-com crash in 2000 or even the 2008 financial crisis.
…

No one could say with any precision how any individual company would profit — indeed, OpenAI itself was burning money at the time and had no clear path to profitability — but everyone knew they had to move fast if they wanted to get in on the action, whatever it ended up being. Google’s chief executive, Sundar Pichai, declared a “code red,” and the company’s co-founders — Larry Page, who was hanging out in Fiji, and Sergey Brin, who had left the company to read physics texts and learn all the Olympic sports — came out of retirement to help. Musk raced to create xAI, his “pro-humanity” A.I. start-up. Zuckerberg directed Meta, the owner of Facebook and Instagram, to launch its first chatbot.
…

This was going to be enormously complicated and capital-intensive, requiring not only expensive microchips but also racks of servers, backup power and extensive cooling systems, not to mention extraordinary amounts of energy and water. Oracle’s hyperscaler rivals were much bigger and better capitalized. If Oracle was going to keep up, it was going to need to borrow a lot of money. And to do that, it was going to need partners with big A.I. ambitions of their own to commit to buying its computing power.

Musk seemed like a perfect fit. He had started OpenAI with Altman in 2015, in what they pitched as a nonprofit effort to develop freely shareable A.I. technology for the good of all mankind. He left a few years later after a bitter power struggle. Musk had since fallen behind in the A.I. race and was now also desperate to keep up.
…

If 2025 was a triumphant year for Ellison, 2026 is shaping up to be very different, as Wall Street seems increasingly anxious about the mountain of debt Oracle has taken on. All of the hyperscalers are making huge investments in A.I. and piling on loads of debt to do it. But Oracle is in a category of its own.

Late last year, as the company borrowed billions of dollars to finance the continuing construction of data centers in Texas, Wisconsin and New Mexico, two credit analysts at Morgan Stanley sent a note to investors estimating that Oracle’s debt and data center lease obligations could triple over the next three years. “Morgan Stanley Thinks You Should Short Oracle,” read a headline in The Financial Times.

Even more striking was Oracle’s so-called debt-to-equity ratio, which stood at around 500 percent — meaning that it had $5 of debt for every $1 of shareholder equity. By comparison, Amazon’s was around 50 percent, and Alphabet’s was considerably lower still.

But Oracle kept borrowing. On a single day in February, the company issued $25 billion worth of bonds. Soon after, it increased its bank credit line to $10 billion, preparing to borrow still more.
Oracle’s debt was becoming an obstacle to its ambitions, as it stretched the limits of the credit market. In March, Oracle was forced to scale back its plans for its Stargate site in Texas after several banks insisted on limiting their commitments to the project because Oracle was the tenant.

That same month, Oracle did what companies do when they are drowning in debt and desperate for cash: It began laying off thousands of employees, roughly 18 percent of its work force, without offering any explanation. It was still unclear if A.I. would decimate the American work force, but it was already decimating Oracle’s. The company’s stock was now collapsing. At the start of April, it was down some 55 percent from its high last September.
…

The story of A.I. has been as much a financial story as a technological one, a question of how to structure the mind-boggling investments required to train and run the models. Few people doubt that this technology is going to change everything. What’s less clear is when the profits are going to start rolling in and how big they are going to be. “To me, it’s a math problem,” says Asad Ramzanali, the director of A.I. at a policy center at Vanderbilt University. “We are making trillions of dollars in investments on the back of tens of billions of dollars in revenues.”

The growing consensus is that these kinds of numbers add up to a bubble. The more salient question may be how big a bubble, and also what will happen if it bursts. One macroeconomic research firm, MacroStrategy Partnership, has estimated that the A.I. bubble is 17 times as large as the dot-com bubble and four times as large as the 2008 housing bubble.

John Cassidy at the New Yorker has a cautionary tale about a young whiz kid Leopold Aschenbrenner who got caught up in the hype and then crashed when lenders, nervous about their money, issued margin calls that required him to sell off many assets at a discount. The question is whether something along these lines might also happen to the bigger companies, with much bigger consequences.

The A.I.-to-the-sky narrative was built on the supposition that OpenAI, Anthropic, and other U.S. tech companies would dominate the industry, and their customers would happily pay high prices for access to proprietary, state-of-the-art models. Early last year, the Chinese A.I. company DeepSeek challenged this narrative by releasing a cheap open-source model with impressive capabilities. That was only the beginning. In the past few weeks, two more Chinese firms, Alibaba and Moonshot, have put out models that, based on some metrics, can match the latest and most powerful offerings from Anthropic and OpenAI. DeepSeek also launched a new model, V4 Flash, which, according to Bloomberg, can execute certain tasks for a cost of three cents, compared with $3.15 when using Anthropic’s Claude Fable 5 model. “When China asks for cents where American companies charge dollars, the contest between the two nations for A.I. customers . . . starts to look materially different,” Bloomberg reported.

AI has become a major election issue as communities across the country are protesting the creation of massive data centers.

About seventy per cent of Americans, including sixty-three per cent of Republicans, oppose building the hyper-scale centers that power the A.I. boom in their area, according to Gallup, but more than four thousand centers have been built across the country, and about four thousand more are planned. Politicians, first from the left and then from the center and the right, have spent the past few months hastily distancing themselves from them. Bernie Sanders and Alexandria Ocasio-Cortez proposed a federal moratorium on new data centers in March, and the pragmatic governor Kathy Hochul enacted the first statewide one in New York in July.
…

Florida’s Republican governor, Ron DeSantis, always on high alert for political opportunity, set conditions on new data centers in his state in May. But the surprise in recent weeks has been a reversal on the issue by business conservatives who not long ago were touting A.I. projects as the cornerstone of a new prosperous age. Just eight months ago, Governor Greg Abbott, at a news conference with the Google C.E.O., Sundar Pichai, bragged, “Texas will be the centerpiece for A.I. data centers for Google in the entire world.” Then, last week, Abbott abruptly announced that all such projects would be suspended, pending an “audit” of their costs and potential benefits. Even the brashest, most pro-tech Republicans are sounding defensive. Vivek Ramaswamy, the former Presidential candidate and doge co-chair, who is now the Republican nominee for governor of Ohio, insisted that “despite my opponent’s claims, I’m not ‘pro-data center.’ ”

These centers are a huge drain on power and water resources which they try to pass off onto the local communities and people are blaming them for the rise in their electricity and water utility bills. Candidates are being pressured to state their opposition to the data centers and many across the political spectrum have done so.

If the AI companies do not start to make a lot of money soon, they may well be in trouble since they cannot continue to borrow such astronomical sums indefinitely.