The stock market suffered a big lurch last Friday, leading some to wonder if it’s on the verge of a more serious collapse. The two main factors were second thoughts about the astronomical valuations of tech companies, and rising inflation.
The tech-heavy NASDAQ fell by about 4.2 percent, dragged down by losses of over 10 percent in major chipmakers, including Micron, Marvell, Intel, AMD, Qualcomm, and Arm Holdings. Nvidia was down more than 6 percent and Broadcom almost 8 percent. The declines erased about $1.2 trillion in market value in a single day. The NASDAQ made back less than a fourth of its Friday losses on Monday.
The soaring stock market has been one of the paradoxes of the Trump economy. The usual explanation is that while consumers feel squeezed by rising inflation, market valuations are high because of the AI revolution and the profits of other tech platform monopolies such as Amazon and Google, as well as those of semiconductor producers.
Friday’s sell-off came against a staggering run-up in the prices of chip manufacturers. Intel stock has increased by 453 percent in the past year. AMD stock is up 303 percent. Both are heavily reliant on the pending AI infrastructure build-out that projects spending in the trillions of dollars.
On Friday, markets were also whipsawed by a whole other collision of good news and bad news. The Labor Department reported better-than-expected job growth, of 172,000 in May. The bond market, fearing more inflation, promptly bid up rates on Treasury securities. Both factors increased the chances that the Fed will raise interest rates, which is always bad news for the stock market.
A close comparison is the dot-com stock bubble and crash of 2000.
Underlying this familiar set of interest-rate dynamics was increasing concern about a possible AI bubble. An absurdly high proportion of the stock market is being driven by AI valuations and by stock prices in other tech platform companies reliant on the AI boom.
The two largest AI companies, OpenAI and Anthropic, are privately held, but both plan to go public sometime in the next several months. The financial press reports that each IPO is likely to be valued at about a trillion dollars. Neither company has formally filed with the SEC, but the OpenAI filing could come any day. OpenAI was valued at $852 billion in a recent private funding round, according to The Wall Street Journal.
One other company in the trillion-dollar club has already launched its IPO. Elon Musk’s SpaceX said in a filing it plans to sell 555,555,555 shares at $135 apiece. That would give it a valuation of around $1.77 trillion, roughly doubling the company’s valuation from six months ago. In 2020, SpaceX was valued at just $36 billion. Its entire profitability is based on monopoly sweetheart contracts with the government. SpaceX trading on the NASDAQ is set to begin on Friday in the largest IPO ever. Assuming that OpenAI and Anthropic move forward with plans for their own IPOs, that adds up to about $4 trillion in just three companies, a massive amount of capital.
ARTIFICIAL INTELLIGENCE MAY WELL LIVE UP to its hype as a transformative technology, but still be wildly overvalued financially right now. A close comparison is the dot-com stock bubble and crash of 2000.
In the late 1990s, the promise of new technology tied to the internet led venture capitalists and other investors to pour money into tech startups. The burn rate of many of these companies far exceeded their profits. Investors were betting on future earnings, and bid up stock values. The money was made from stock plays, not from actual profits. The tech-heavy NASDAQ nearly tripled in value between 1997 and early 2000.
When several startups failed, and the Fed began raising interest rates in 1999, the herd instinct reversed and investors began selling. In the crash that followed, the NASDAQ lost 78 percent of its value between March 2000 and October 2002. The crash did not immediately collapse the larger economy. That would come later, with the housing bubble collapse in 2008. Some tech companies, with viable business models, recovered and were soon bigger than ever, such as Amazon, Google, and eBay.
The AI analogy is imperfect but powerful. Like the tech bubble stocks of 1999 and 2000, AI’s rate of capital use and run-up in its projected value far exceed its revenue. A great many users of OpenAI’s ChatGPT and Anthropic’s Claude use it for free. The devil’s bargain that consumers make is that they get to use these applications at no cost, and in exchange the companies can use their data for AI training and God knows what else.
The AI companies treat these free applications as gateway drugs and try to convince consumers, especially business consumers, to shift to paid applications. This is surely coming, but not necessarily at a pace sufficiently fast to justify the astronomical valuations of AI and AI-related companies.
Most of the trillions of dollars that AI companies are raising and spending is projected to go to the build-out of data centers. OpenAI has committed around $1.4 trillion to data center infrastructure over the next eight years and doesn’t expect to have positive free cash flow until 2030. Nvidia CEO Jensen Huang recently said that total AI infrastructure spending could total $3 trillion to $4 trillion by 2030.
Some of the smartest people in the world are trying to figure out how to monetize all of that capital outlay. They are only partly succeeding.
In 2024, David Cahn of Sequoia Capital, one of the most influential AI bears, published a much-quoted report posing what he called AI’s $600 billion question. The $600 billion was the revenue needed to justify AI’s immense capital expenditures. He was skeptical that AI would earn anything like enough.
Since then, Cahn has conceded that the revenue of AI companies is growing. OpenAI reported revenue of $20 billion in 2025. But part of this growth is coming from the repricing of AI for business consumers, which has led to astronomical costs, when businesses were promised savings from firing workers and replacing them with technology. One company reportedly spent $500 million on Claude in a single month, after Anthropic shifted to usage-based billing. If businesses won’t take on these costs, AI firms could lose their only reliable potential revenue growth.
The equation is further complicated by a broad-based and growing popular revolt against AI data centers, which are raising the cost of electric power for other users and stressing the grid.
Illinois Gov. JB Pritzker announced that starting July 1, his administration will stop approving new state tax breaks for data centers in Illinois. He has ordered the state Commerce Department to cease processing new applications and has asked lawmakers, utilities, labor, and environmental groups to come up with a whole new approach.
Pritzker wants tech companies to pay their own way, with a separate rate class for data centers. He wants them to fund their own clean energy instead of making ratepayers finance their AI boom. He wants their power cut first when the grid is strained.
In Virginia, as my colleague Gabrielle Gurley reports in this Prospect piece, one county after another has rebelled against data centers. In Republican Utah, a plan to create a massive data center in Box Elder County on 40,000 acres (more than twice the size of Manhattan) has produced growing citizen backlash. The project will require about nine gigawatts of power, or more than the entire state of Utah currently consumes. It will also use huge amounts of water in a county that has suffered severe drought in recent years. In response, the project’s sponsor, celebrity investor Kevin O’Leary, has offered to cut the scale to 10,000 acres.
The pushback against data centers has barely begun, and as more limits are imposed, that will also raise questions about the inflated valuation of AI companies and their burn rate. A national debate about how to regulate AI is just beginning.
Meanwhile, Bernie Sanders had a rare bad idea. Sanders proposed nationalizing part of AI. Some AI execs as well as President Trump jumped on the bandwagon. But the devil is in the details. Bringing in government as a partner at the peak of a bubble will only inflate the bubble, and leave taxpayers holding the bag for a bailout on the downside. And as our friend Matt Stoller asks, if we are going to give government a stake in explicit AI companies like OpenAI and Anthropic, what about part public ownership of more diversified tech monopolies like Google that are heavily invested in AI and have their own AI products.
FINALLY, LET’S TAKE A CLOSER LOOK at that good jobs report, which set off Friday’s panic selling in financial markets. Despite Trump’s boasts about the economy (“It’s raining jobs!”), the job creation is not translating into wage growth.
Average hourly earnings for private-sector workers rose by only 0.2 percent in May, an annual rate of 2.4 percent. But prices are rising at almost double that rate, around 3.8 percent annually. So real wages are falling.
Even so, why is this relatively flabby economy producing jobs at this rate? The answer is that the federal deficit is now about $1.9 trillion, or about 5.9 percent of GDP. That’s close to double the 50-year average. In other words, the job creation is the result of pure Keynesian stimulus.
Unfortunately, it’s the wrong kind of Keynesian stimulus. It is driven almost entirely by tax cuts for the wealthy rather than public investment, which has been cut. And it’s not countercyclical, but it’s year in and year out, which means that it’s unsustainable. As the Fed raises interest rates, the cost of carrying all that debt will become its own drag on the economy.
There was a time when Republicans, as fiscal conservatives, might have objected. But the Republican Congress keeps on rubber-stamping Trump’s budget deficits with nary a word of protest.
The tax cuts, which put more money into the pockets of wealthy investors, also interact nicely with the stock market bubble. All of which suggests that when the crash does come, it will be a doozy.
