What Is the AI Bubble? Signs, Risks and What Happens If It Pops
Is AI a bubble? Here is what circular financing, the spending-revenue gap, market concentration and the Fed are actually saying right now.
The "AI bubble" is the name for a running argument about whether the trillions of dollars pouring into artificial intelligence are rational investment in a transformative technology or a speculative mania that is due to deflate, Wall Street-style. There is no single event that would prove either side right overnight. Instead, the debate turns on a specific set of financial warning signs — circular deals between a handful of companies, spending that dwarfs actual AI revenue, and a stock market more concentrated in a few names than at any point in half a century — weighed against counter-arguments from the companies doing the spending. Here's what the term means, the evidence on both sides, and what a "pop" would and wouldn't change.
Quick facts
- "AI bubble" describes fears that AI-related valuations and spending have outrun the technology's current revenue and productivity gains.
- Warning signs include circular financing between AI labs and chipmakers, data-center capex estimated near $3 trillion for 2025–2028, and the top five S&P 500 companies holding roughly 30% of the index's weight.
- Nvidia CEO Jensen Huang, Fed Chair Jerome Powell, and several Wall Street banks have pushed back on the bubble framing, arguing AI firms generate real revenue unlike many dot-com-era startups.
- The Federal Reserve's November 2026 Financial Stability Report named AI financing as the top risk cited by market contacts, the first time it devoted a dedicated section to AI.
What Does "AI Bubble" Actually Mean?
An economic bubble forms when the price of an asset — a stock, a currency, a type of property — rises well beyond what its underlying earnings or utility can justify, usually fueled by borrowed money and the expectation that someone else will pay even more for it later. The "AI bubble" argument applies that pattern to the current buildout of AI infrastructure: the data centers, chips, and cloud capacity that companies like OpenAI, Anthropic, Microsoft, Google, Meta, and Nvidia are committing hundreds of billions of dollars toward.
The debate isn't about whether AI itself is useful. It's about whether the current pace and structure of investment in AI — who is funding it, how much revenue it is actually producing today, and what happens to credit markets and stock valuations if growth disappoints — looks more like durable infrastructure-building or more like a speculative peak.
The Case That There Is a Bubble
By 2026, AI infrastructure spending had shifted from being funded mostly out of companies' own cash flow to being increasingly financed through corporate bonds and an estimated $3 trillion in off-balance-sheet commitments, according to Wikipedia's running summary of the debate. Critics point to several specific patterns:
- Circular financing. Leading AI companies are investing in each other in ways that can inflate the group's combined valuation without new outside money entering the system.
- Spending that outpaces revenue. Several of the biggest AI buildouts commit far more in multi-year infrastructure spending than the companies currently earn in revenue.
- Record market concentration. A small number of AI-linked stocks account for an unusually large share of major indexes.
- Debt growing faster than oversight. Regulators have flagged the pace of AI-linked borrowing as a new category of systemic risk.
Circular Deals, Explained
The clearest example of circular financing came in September 2025, when Nvidia announced a $100 billion investment in OpenAI that was tied to OpenAI purchasing Nvidia GPUs with that same money — a deal the Wall Street Journal described as a structure that "can be virtuous on the upside, and vicious on the downside." The following month, OpenAI struck a large hardware deal with AMD that included AMD handing OpenAI a major equity stake, and Oracle signed a $300 billion compute-supply agreement with OpenAI. Each deal, on its own, is a normal commercial transaction. Critics argue that stacked together, they let a handful of companies effectively fund their own customers' demand for their products — and that the resulting revenue and valuation gains can look stronger than the underlying cash economics actually are.
The Numbers: Spending vs. Revenue
The gap between committed spending and current revenue is the single most-cited data point in the bubble argument. OpenAI has reportedly committed roughly $1.4 trillion over eight years to data-center capacity, against revenue far smaller than that figure, and has projected annual operating losses reaching an estimated $74 billion in 2028 alone. Morgan Stanley has estimated global data-center capital spending at around $3 trillion for 2025–2028, with roughly half of that financed through private credit rather than company cash.
| Metric | Figure | Why it matters |
|---|---|---|
| Global data-center capex, 2025–2028 (est.) | ~$3 trillion | Scale of the bet relative to any single company's balance sheet |
| OpenAI data-center commitments | ~$1.4 trillion over 8 years | Multi-year spending far exceeding current revenue |
| OpenAI projected 2028 operating loss (est.) | ~$74 billion | Illustrates the revenue-spending gap critics point to |
| AI-linked public debt outstanding (Oct. 2026) | >$1.1 trillion | How much of the buildout rests on borrowed money |
| Top-5 S&P 500 companies' index weight (late 2025) | ~30% | Highest market concentration in roughly 50 years |
Big Tech earnings have already shown markets reacting to this gap: after one quarter in which Microsoft spent close to $35 billion on AI infrastructure, its shares fell even though revenue grew 18%, as investors focused on the cost of the buildout rather than the growth it produced.
How Concentrated Are Markets?
Heading into 2026, Nvidia, Microsoft, Apple, Alphabet, and Amazon together accounted for close to 30% of the S&P 500's total weight and around 20% of the MSCI World index — a concentration level not seen in roughly half a century. The S&P 500 was trading at around 23 times forward earnings, and the Shiller P/E ratio for the broader U.S. market climbed above 40 for the first time since the dot-com crash. AI-related companies were estimated to account for roughly 80% of total U.S. stock market gains in 2025. None of that proves a bubble by itself, but it does mean that if AI-linked valuations fall sharply, the effect on broad market indexes — and on retirement accounts tracking them — would be larger than a typical sector pullback.
The Case Against a Bubble
The companies at the center of the spending reject the bubble framing. Nvidia CEO Jensen Huang has repeatedly dismissed bubble talk, telling audiences at the World Economic Forum in Davos in January 2026 that AI represents the largest infrastructure buildout in human history rather than a speculative peak, and arguing on a later earnings call that Nvidia sees "something very different" underway as more computing shifts to accelerated, AI-ready hardware.
Federal Reserve Chair Jerome Powell has made a similar distinction, arguing that AI differs from the dot-com era because today's leading AI companies generate substantial real revenue, unlike many money-losing dot-com startups that had little more than a business plan. Goldman Sachs and Morgan Stanley analysts have likewise argued that current valuations, while elevated, remain below dot-com-era extremes given how much actual revenue today's AI leaders produce.
What the Federal Reserve Is Watching
The clearest sign that regulators now take the debate seriously came in the Federal Reserve's November 2026 Financial Stability Report, which devoted a dedicated section to AI financing for the first time. In the Fed's survey of market contacts, 87% named AI as a top risk to financial stability, up from 50% in the spring 2026 survey. The report noted that equity valuations "have remained high, partly driven by market enthusiasm for super intelligence," while also finding that hedge-fund leverage had reached a record level in data going back to 2013. Fed officials are not unanimous: New York Fed President John Williams has said he does "not see this as a bubble kind of situation," while Governor Lisa Cook and others have been more cautious, noting that many of the AI spending commitments announced so far are still pledges rather than built, operating infrastructure — which, for now, limits the risk of those assets sitting stranded if demand falls short.
Who Would Be Hit Hardest?
Not every company sitting near the word "AI" carries the same risk. The companies most exposed to a sharp correction are the ones furthest from having actual AI product revenue today: smaller data-center builders and power-infrastructure firms that took on debt betting on multi-year lease commitments from a handful of AI labs, private-credit funds that financed much of that roughly $3 trillion data-center buildout outside the regulated banking system, and chip-supply-chain companies in markets like South Korea and Taiwan whose stocks rose almost entirely on AI-demand expectations. The companies with the most cushion are the ones that already had large, profitable businesses before AI became the dominant story — Nvidia's chip sales, Microsoft's cloud and software subscriptions, Alphabet's advertising business, and Amazon's retail and cloud operations all generate substantial non-AI revenue that would keep running even if AI-specific spending slowed sharply. That split is also why economists at the Fed have been careful to separate "AI is overhyped" from "an AI-linked credit event could still spread through the wider financial system" — the second risk doesn't require the first to be true.
What Happens If It Pops?
Historical comparisons point both ways. Ray Dalio has said AI investment levels look "very similar" to the dot-com bubble, and analyst Julien Garran has estimated the current AI buildout at roughly 17 times the size of the dot-com bubble by some measures. The 2000 dot-com crash wiped out companies with no revenue model at all; the Nasdaq fell about 78% from its March 2000 peak by October 2002. A comparable unwind in AI-linked stocks today would hit far more of the broader market, given how concentrated major indexes have become in a handful of AI-exposed names — but unlike many dot-com casualties, today's largest AI companies have real product revenue, real customers, and (in Nvidia's and the hyperscalers' cases) profitable non-AI businesses to fall back on. A pop would most likely mean a sharp repricing of AI-linked stocks and tighter credit for the smaller, more leveraged players building data centers on borrowed money — not the disappearance of AI products people already use daily. That pattern echoed in July 2026, when South Korea's KOSPI index fell roughly 44% in 40 days amid fading enthusiasm for the country's AI-linked chip boom, a preview some analysts cite as a smaller-scale version of what a broader unwind could look like.
What Next
Watch three things in the coming months: whether AI companies' actual revenue growth starts closing the gap with their spending commitments, whether credit spreads on AI-linked corporate debt keep widening (Oracle's five-year credit-default-swap spread hit a record in October 2026), and what the Fed's next Financial Stability Report says once more of today's announced data-center spending has actually been built rather than just promised. None of those will settle the argument cleanly — bubbles are usually only confirmed in hindsight — but they're the clearest signals available right now for whether the AI buildout is heading toward a soft landing or a sharp correction. For more on the financial side of this story, see Pandromeda's coverage of Anthropic's IPO pricing and the Claude vs. GPT race that is driving part of this spending in the first place.
Frequently asked questions
Is the AI bubble going to burst?
No one knows for certain — bubbles are usually only confirmed in hindsight. The Federal Reserve's November 2026 Financial Stability Report flagged AI financing as the top risk cited by market contacts, but Fed officials are themselves divided, and companies like Nvidia continue to post real, growing revenue rather than just rising stock prices.
What is circular AI financing?
It refers to deals where AI companies and their suppliers invest in each other, such as Nvidia's $100 billion investment in OpenAI tied to OpenAI buying Nvidia GPUs. Critics say these loops can inflate the group's combined valuation without new outside capital entering the system.
How is the AI boom different from the dot-com bubble?
Federal Reserve Chair Jerome Powell and several Wall Street banks argue today's leading AI companies generate substantial real revenue, unlike many dot-com-era startups that had little more than a business plan. Critics counter that spending still far outpaces that revenue in dollar terms.
Who would be hurt most if the AI bubble pops?
Smaller, more leveraged players are most exposed: data-center builders and power-infrastructure firms financed by private credit, and chip-supply-chain companies whose stocks rose mainly on AI-demand expectations. Companies with large non-AI revenue, like Microsoft's cloud/software business or Amazon's retail operations, have more cushion.
What is the Federal Reserve doing about AI financial risk?
The Fed's November 2026 Financial Stability Report devoted a dedicated section to AI financing for the first time, surveying market contacts (87% named AI a top risk) and tracking hedge-fund leverage and AI-linked corporate debt, which topped $1.1 trillion outstanding by October 2026.
Sources
- Federal Reserve: Financial Stability Report (Nov. 2026)federalreserve.gov
- Wikipedia: AI bubbleen.wikipedia.org
- Wikipedia: Dot-com bubbleen.wikipedia.org
- Nvidia Investor Relations: Quarterly Resultsinvestor.nvidia.com
Theo Park runs the AI desk at Pandromeda. He follows model launches from the frontier labs and the open-weight community, tracks the assistants and developer tools built on them, and explains what each release changes on pricing, capability and safety. His reporting leans on primary sources: model cards, technical reports, API documentation and the companies' own announcements.


