Is the AI Bubble Bursting? The 10 Warning Signs, Explained

Something has shifted in how the world talks about artificial intelligence. Search interest in the “AI bubble” spiked into the UK’s top-10 trending searches this week, the IMF, the Bank of England and the Federal Reserve have all flagged AI valuations, and Gary Marcus is comparing the current moment to 2008. At the same time, Big Tech keeps spending — Alphabet raised its 2026 capital expenditure range to $195–205 billion in July — and Nvidia’s revenue just jumped 106%.

In brief

• Is the AI bubble bursting? The IMF’s warning, Gary Marcus’s 2008 comparison, and the 10 warning signs economists and central banks are watching.
• 1. Big Tech keeps raising AI spending, not cutting it: The most bullish sign for the boom — and the reason bubble-callers haven't won yet — is that spending hasn't slowed.
• 2. Nvidia's growth shows no cracks: The chipmaker at the center of the boom posted revenue up 106% to $96.2 billion last quarter, with $108 billion forecast for the next.
• 3. Debt stress is showing at the weaker players
• 4. The IMF, Bank of England and Fed are all flagging it: All three institutions have publicly warned about AI-driven valuations and the risks in government bond markets.

So is the bubble popping, or is this a healthy correction? Here are the 10 warning signs economists and investors are watching right now, and where each one actually stands.

Why this particular week matters: it is unusual for the IMF, the Bank of England and the Federal Reserve to sound the same note within days, and unusual for a single phrase — “AI bubble” — to climb national trending lists. When central bankers and search interest agree, the argument has moved from trading desks to dinner tables.

1. Big Tech keeps raising AI spending, not cutting it

The most bullish sign for the boom — and the reason bubble-callers haven’t won yet — is that spending hasn’t slowed. Alphabet lifted its 2026 capex range upward in July; Meta narrowed its range upward to $130–145 billion. If this were a classic investment bubble peaking, you’d expect the biggest spenders to blink first. They haven’t.

Counterpoint: Critics note that relentless spending is exactly what happened in 1999 — the money flows until it can’t.

2. Nvidia’s growth shows no cracks

The chipmaker at the center of the boom posted revenue up 106% to $96.2 billion last quarter, with $108 billion forecast for the next. One widely cited warning-sign checklist puts it plainly: Nvidia’s growth would have to stall for the bubble thesis to hold. It hasn’t stalled — demand for AI chips is still accelerating.

Counterpoint: Nvidia is a beneficiary of the boom, not a barometer of its sustainability. Dot-com infrastructure suppliers also boomed before 2000.

3. Debt stress is showing at the weaker players

Here’s where the red lights start. Bond yields at CoreWeave and Oracle hit records in October 2026, according to market data cited in recent bubble analyses. Data-center builders are among the most leveraged players in the AI trade, and the bond market is charging them record prices to borrow.

On October 14–15 the pressure showed in share prices too: CoreWeave fell 15% in a day and Oracle nearly 9%, after hotter-than-expected US inflation data pushed traders to bet on more interest-rate rises. And the scale of the borrowing is new, not just the price: the Bank of England’s Financial Policy Committee cited Morgan Stanley estimates of roughly $450 billion in AI-related debt issued worldwide by early September 2026 — more than double the whole of 2025.

4. The IMF, Bank of England and Fed are all flagging it

All three institutions have publicly warned about AI-driven valuations and the risks in government bond markets. IMF chief Kristalina Georgieva warned in Bangkok that “if there are no safeguards, then the risks may exceed the opportunities.” When the world’s central bankers and the IMF say it in the same week, markets listen.

5. Gary Marcus warns of a 2008-style crisis

AI critic Gary Marcus told The Economic Times on October 17 that a sharp reversal in AI investments could trigger a financial crisis resembling 2008. His argument: high valuations, extensive borrowing, and growing financial links around AI infrastructure mean losses at one major company could spread through markets. He specifically questioned whether OpenAI can justify its spending commitments and sustain the valuation expectations around its future public listing — comparing its trajectory to WeWork, whose planned 2019 IPO collapsed before bankruptcy in 2023.

Qualification (his own interviewers’ caveat): A market correction would not automatically produce a banking crisis. The consequences would depend on leverage, creditor exposure, and how financial institutions respond.

6. $1.4 trillion of circular deals

The IMF counts $1.4 trillion in revenue promises tied to circular AI deals — AI firms effectively funding each other. None has unwound yet, but the structure means that if one link breaks, the chain can tighten fast. This is the mechanism behind the contagion argument: lenders, bondholders, data-center operators and other investors all sit downstream of the same spending.

The IMF has since put numbers on the structure: its review of 14 deals struck since 2023 between cloud giants and AI firms found roughly $300 billion in equity investments and $650 billion in debt, given in exchange for the $1.4 trillion in future revenue commitments. If one key company ran into trouble, the IMF warns, revenue forecasts and valuations would fall, weakening the value of the debt accumulated in those deals.

7. AI revenue is growing — but so are the losses

Revenue is growing fast at the big labs, but compute bills and losses are growing alongside it. The question investors are now demanding answers to during earnings season: is AI revenue catching up with AI spending? The Nasdaq finished the week of October 16 down 4.5%, even though it hit a record in between sharp falls in July and mid-October. The era of unquestioned AI enthusiasm on Wall Street appears to be over — from here, the market wants receipts.

8. China issued policy guidance warning against AI investment bubbles

On October 17, China published policy guidelines urging stronger technological innovation while warning against speculative investment and poorly coordinated expansion in emerging sectors. When the world’s second-largest economy formally tells local governments to avoid wasteful duplication and financial bubbles in AI, it’s a policy-level confirmation that bubble fears have gone mainstream — even inside state planning systems.

9. Interest rates could puncture it

Traders are betting on more US rate rises, and rising rates make borrowing to build data centers more expensive. Moody’s Analytics economists added that a prolonged Middle East conflict — pushing up energy prices and interest rates — could “puncture” the AI boom by straining the economics behind AI investment. Three signposts matter in the coming weeks: the Fed’s October 28 policy decision, ongoing earnings season, and confidence itself.

10. AI stocks are wobbling — and Anthropic’s IPO has no date

AI stocks fell sharply in July and again in mid-October, though they recovered in between. Anthropic’s long-discussed IPO still has no date and has reportedly slipped once. If big listings get delayed or priced lower, that’s traditionally one of the last signals before a bubble breaks.

The stakes of that listing are concrete: Anthropic’s IPO paperwork reportedly shows around $518 billion of compute commitments, roughly 80% of them non-cancellable — the kind of fixed obligation that makes a listing’s pricing a referendum on the whole boom. A listing at or above the hoped-for $2 trillion valuation would signal investors’ appetite is intact; a delay or a cut price would not.

What both sides agree on

• Nobody disputes the valuations are extreme. The argument is about whether the spending behind them is productive.
• Nobody disputes demand for AI chips and data centers is real. The argument is about whether the financing behind them is sustainable.
• The data-center builders (Clayco’s Bob Clark among them) insist demand is unprecedented and spending is far from over.

A note on how to read these headlines

Two different arguments are wearing the same “bubble” costume, and keeping them apart makes the coverage clearer. One is about valuations: whether stocks and private-market prices have run ahead of what the businesses will actually earn. The other is about technology: whether AI systems deliver enough value to justify the infrastructure bill. The valuation argument can be right while the technology argument is wrong — you can believe AI will transform industries and still think the shares are priced for perfection. Almost every serious warning this month is the first kind, not the second. Watch for commentators who quietly swap between them; a bubble in prices is not the same thing as a verdict on the technology.

What to watch next

The Fed’s October 28 decision lands a week before the US midterms. Then earnings season will show whether AI revenue is finally catching up with AI spending. Until then, the honest answer to “is the AI bubble bursting?” is: the warning lights are on, but the two biggest — Big Tech spending and Nvidia’s growth — are still green.

(This article is an explainer of publicly reported warnings and market data. It is not financial advice.)

Sources

• Startup Fortune — IMF warns an AI stock correction could threaten stability
• FinTech Global — Bank of England and IMF heads warn on AI and cyber risk

Related reading: Microsoft Pulls Out of Nscale UK Data Center Deal: Why It Matters · NVIDIA’s RTX Spark laptops and what the AI-chip boom means for buyers · PlayStation 6: Sony’s three models and what they mean

Update, October 21, 2026: one more data point for the AI-economics side of this story — a single hacker spent roughly $5,000 in AI tokens, about $50 per successful intrusion, to breach more than 100 company websites with AI agents. When autonomous capability costs less than a business lunch, the economics of AI are doing more than inflating valuations — they are rewriting the cost model of attack.

Update, October 21, 2026: a fresh data point on the cost of autonomy — DeepMind’s four AI agents needed 27 hours, 2,783 programs, and about $520 in tokens to beat Minecraft’s Ender Dragon with no human help. Genuine long-horizon capability, and a blunt reminder of what the token bill looks like for it.