A few days ago Anthropic and OpenAI filed confidential S‑1 documents with the U.S. Securities and Exchange Commission (SEC). Anthropic currently runs at a $965 billion valuation, while OpenAI is valued at $850 billion. These eye‑watering figures have prompted investors to ask whether they reflect durable value or the final, most conspicuous moments before a large bubble pops.
The companies’ financial profiles are mixed. OpenAI’s revenue grew from roughly $200 million in 2022 to more than $10 billion by 2025, yet profits are absent: the company reported a $13.5 billion net loss in the first half of 2025 and expects about a $14 billion loss in 2026. It is burning roughly $17 billion in cash this year. Internal documents foresee $44 billion of cumulative losses between 2023 and 2028, with profitability pushed out to 2029. Burning multibillions annually while seeking trillion‑dollar valuations looks disproportionate.
Funding structure: circular flows of capital
What differentiates this episode from a classic overheated market is the financing structure. Key industry players increasingly invest in each other, creating a circular flow of capital. Nvidia announced a $100 billion partnership with OpenAI, but much of that funding will be spent by OpenAI on buying Nvidia chips. Microsoft owns about 27 percent of OpenAI while also serving as its main cloud provider, so Azure revenues can cycle back into hardware purchases. Nvidia also owns 7 percent of CoreWeave and agreed to buy $6.3 billion of idle, GPU‑stocked data‑center capacity from CoreWeave. The danger is clear: some of today’s high valuations are sustained by participants inflating each other’s revenues and prospects rather than broad external demand.
Stock market concentration
The risk is not confined to private markets. U.S. stock indices are being driven higher largely by the AI narrative, and concentration has reached unprecedented levels. The “Magnificent 7” account for 35–40 percent of the S&P 500’s market value, and Goldman Sachs estimates that these seven stocks contributed 46 percent of the index’s 2026 growth. That implies average investors—via index funds—are betting on AI success to a greater degree than they may realize.
At the same time, an often‑cited MIT study finds that 95 percent of organizations investing in generative AI currently see zero returns. There remains a substantial gap between technological promise and realized corporate benefits.
Historical parallels and warnings
Pessimists have drawn parallels to the 1999–2000 dot‑com bubble. Michael Burry, known for forecasting the 2008 housing collapse, said the market feels like “the last months of the 1999–2000 bubble” and placed large bets against Nvidia and Palantir. According to Burry, 87 percent of venture capital, 49 percent of investment‑grade bond issuance, and 38 percent of high‑yield debt are today tied to AI. Goldman Sachs says today’s AI boom resembles conditions in 1997—years before a potential bust—and lists five warning signs: peak capital expenditures, falling corporate profits, rising corporate debt, rate cuts, and widening spreads.
Why a collapse could hurt the whole economy
A particularly troubling element is that AI already significantly affects real economic growth. The article reports that AI‑related investment added 1.1 percentage points to U.S. GDP growth in the first half of 2025, outpacing consumption as the growth driver. Harvard economist Jason Furman goes further, estimating that AI infrastructure investment accounted for 92 percent of U.S. GDP growth in that period. In practice, much of the apparent economic strength stems from a handful of hyperscalers—Microsoft, Google, Amazon, Meta and Oracle—rushing to build data centers. The nine largest cloud providers plan roughly $830 billion of investment by 2026, equal to about 2.2 percent of U.S. GDP.
This growth is fragile because capital expenditure can be paused quickly. If investor confidence wavers and funding dries up, companies can delay or cancel data‑center plans, triggering falling stock prices, reduced household wealth, weaker consumption and lower corporate profits — a self‑reinforcing downward spiral. Analysts at Oliver Wyman and the World Economic Forum warn that such a collapse could cause a deeper recession than the 2000 dot‑com crash, because more U.S. households now own equities and would therefore suffer larger wealth effects. The shock would also reverberate globally: U.S. market corrections typically spill into Europe and Asia, and export economies exposed to chips and data‑center supply chains—from Taiwan to South Korea—would be directly affected.
A moderating factor
There is a mitigating factor: the bubble is unusually concentrated among relatively few companies, so the direct effects on consumption and employment may be more limited than in a broad‑based collapse. The biggest losers would likely be speculative players that rely on cheap capital and lack profits.
Interpreting the bubble label
It is important to separate market exuberance from the technology’s intrinsic value. Bubbles form not because a technology is useless, but because markets overshoot its near‑term prospects. The dot‑com episode shows the internet ultimately transformed the world, even though many early firms failed. Proponents argue that AI infrastructure—chips, data centers and the energy that supports them—constitutes tangible, durable capital that will outlast the hype, and that today’s rally is driven by some of the world’s most profitable companies with diversified businesses. Moreover, demand for certain AI solutions is real, with paying customers and rapidly growing revenues, even if returns lag investment.
Conclusion: uncertainty with material risk
The truth lies between “everything collapses” and “nothing to worry about.” The market almost certainly contains bubble elements: hundreds of billions in mutual commitments, historic stock concentration and multibillion dollar cash burn without profits all point to overheating. Yet the underlying technology and infrastructure may have lasting value, and large, profitable firms are funding much of the experimentation.
No one can say reliably when or how the excess air will leave this bubble. Some forecasts predict a gradual slowdown or sharp correction as early as 2026; others tie the turning point to interest rates, inflation and corporate profits. Historically, bubbles persist longer than intuition suggests and then burst faster than most can prepare for. One thing seems clear: if this bubble pops, the pain would extend far beyond Silicon Valley because much of the U.S. growth story—and global market sentiment—today rests on the AI narrative.
