A prediction market carrying $2.9M in trades now prices the odds of a full AI bubble burst by the end of 2026 at just 17%. Traders are betting against a collapse that central bankers, in the same season, are actively warning about. The Bank for International Settlements used its 2026 annual report to flag something new: AI infrastructure is increasingly financed with debt, credit spreads on that debt are widening, and circular financing arrangements between chipmakers, cloud providers, and AI labs could carry a shock through the financial system rather than contain it in Silicon Valley. That warning lands seven months after this analysis first argued the spending-to-revenue gap was unsustainable. The gap has not closed. What has changed is how it's being paid for.
The Spending Figure Nobody Predicted Correctly
Every forecast made in late 2025 about 2026 AI spending has already been beaten. Gartner now projects global AI spending will cross $2.52 trillion in 2026 — a figure that dwarfs the $725 billion hyperscaler capex number that anchored the first version of this piece, and one that JPMorgan says could be followed by another $5 trillion in AI infrastructure spending over the next four years from the four largest cloud platforms alone. Enterprise AI revenue has grown too, but nowhere near that pace — most 2026 estimates still cluster near $100 billion once vendor self-reporting is stripped out.
The ratio between what's being built and what's being paid for hasn't narrowed since November 2025. It has widened. And for the first time, the companies footing the bill are starting to say so out loud, in the careful, lawyered language of an earnings call rather than the plain arithmetic a shareholder actually needs.
What Dot-Com and 2008 Still Get Half-Right
Reach for the obvious historical comparison and it only half fits. The dot-com crash wiped out $4.4 trillion because companies with no revenue were priced as if revenue were inevitable. The 2008 crisis destroyed $17 trillion because a debt instrument nobody understood was quietly load-bearing for the entire financial system. The AI bubble borrows a symptom from each — inflated valuations from 2000, and now, increasingly, hidden debt exposure from 2008 — without being a clean copy of either.
| Crisis | Trigger | Losses | Recovery Time |
|---|---|---|---|
| Dot-Com Bubble | Internet equity speculation | $4.4 trillion | ~7–9 years |
| Mortgage Crisis | Subprime debt securitization | $17 trillion | ~5–6 years |
| AI Bubble (projected) | Infrastructure overbuild + debt-financed capex | Up to $70 trillion | Unknown — no precedent |
The $70 trillion exposure figure isn't a single company's market cap. It's what happens when a correction hits cloud infrastructure, semiconductors, energy, commercial real estate, and public equities at the same time — sectors that, unlike 2000's internet-only carnage, are now all leveraged to the same technology bet simultaneously.
The Debt Signal Nobody Priced In
The original version of this analysis noted, accurately at the time, that hyperscalers were funding AI capex almost entirely from earnings — a genuine point in the bull case. That's no longer the full picture. The BIS's 2026 annual report warns that AI capex is increasingly financed with debt, that credit spreads tied to that debt are widening, and that circular investment structures — where chipmakers fund the customers who buy their chips — raise the odds that a slowdown spreads through the credit system instead of staying contained to equity valuations.
Moody's data backs the shift up in dollar terms: tech companies issued $108.7 billion in corporate bonds in a single quarter at the end of 2025, and that pace has held through the first half of 2026.
"It's a lot of debt, and a lot of it all of a sudden.
Mark Zandi, Chief Economist, Moody's AnalyticsThis is the mind-changing part of the story. A year ago, the earnings-funded thesis was the strongest argument against a systemic AI collapse. It hasn't been disproven — but it's being quietly eroded, one bond issuance at a time, by the same companies that used it as their defense.
📌 Related Reading: why autonomous AI agents change the economic stakes of this entire build-out — the infrastructure being financed with debt today is largely being built for systems that act, not just answer.
What the Market Is Actually Pricing In for an AI Bubble Burst
Polymarket's "AI bubble burst by...?" contract has moved $2.9 million in volume since it opened in November 2025. The single most-backed outcome isn't "never" — it's December 31, 2026, the last date on the board before a full year of earnings reports has to reconcile the spending with the revenue. That's not certainty. It's a crowd hedging against the deadline it can't see past.
Forrester has gone further, calling for an outright AI market correction in 2026, pointing out that fewer than a third of enterprise decision-makers can tie AI spending to measurable financial return — which means the CFOs who approve 2027 budgets will be doing so with far less patience than the CTOs who approved 2024's.
Anatomy of a Break: Three Cascade Scenarios
The Equity Scenario
US equity market capitalization sits near twice GDP — well above the dot-com peak. A confidence shock in AI valuations would flow directly through ETFs, pension funds, and household portfolios, where stock ownership now represents roughly 30% of total wealth.
The Hybrid Scenario (Debt-Accelerated)
This is the scenario the BIS warning makes more likely, not less. When capex is funded by debt instead of cash flow, a revenue shortfall doesn't produce a slowdown. It produces a margin call.
The Double Unemployment Crisis
McKinsey's 2025 projections put potential global job displacement from AI and automation at 400 to 800 million workers by 2030. A financial-side collapse layered on top of that displacement would be historically unusual: the technology causing the downturn is the same one eliminating the jobs that would normally cushion it.
The Case Investors Keep Making — And Its Limits
The bull case hasn't disappeared, and it deserves a fair hearing. Microsoft, Alphabet, Meta, and Amazon generate real profits, in the hundreds of billions, and their AI chip purchases come from paying enterprise customers rather than speculative startups. Goldman Sachs and J.P. Morgan continue to describe the buildout as fundamentally justified rather than speculative mania.
Michael Burry — who built his reputation correctly calling the 2008 crash before almost anyone believed him — is on the other side of that argument, publicly positioning against an AI bubble he expects to burst, and pointing to declining returns on invested capital across Big Tech's AI spend as evidence the market is ignoring.
Neither side disputes that AI is a real technology with real long-term value. The disagreement is about timing: whether 2026's prices already assume a productivity payoff that's still years away.
📌 Related Reading: how export controls on frontier AI models are reshaping who controls this infrastructure — a regulatory wrinkle that adds another variable to how contained or global any AI-sector correction would be.
What Investors and Policymakers Should Actually Do
Avoiding AI exposure entirely is its own risk — missing a real structural shift rarely ages well. The more useful question is where in the value chain exposure sits, and how it's financed. Companies with positive cash flow, defensible moats, and AI spending that's supplementary rather than existential carry a fundamentally different risk profile than companies whose entire valuation rests on AI promise alone. Energy, power transmission, and cooling infrastructure may hold up better than model-layer companies whose pricing power keeps getting undercut by open-source releases within weeks of launch.
For policymakers, the BIS warning is the clearest signal yet that disclosure requirements around AI-linked debt and circular financing arrangements need to move faster than the spending itself. The 2008 crisis showed what happens when systemic risk gets identified after the fact rather than before it.
Who This Is For
This matters most for anyone with direct exposure to AI-adjacent equities or bonds — retail investors holding concentrated tech positions, portfolio managers weighing 2027 allocations, and policymakers drafting the disclosure rules the BIS is now implicitly asking for.
The Verdict
The spending-revenue gap that made the original case for an AI bubble hasn't closed — it's widened, and it's now partly debt-financed, which raises the stakes of any correction rather than lowering them. That doesn't mean AI is a mirage; it means current prices are betting on a payoff timeline the data doesn't yet support. Selective exposure, weighted toward companies with real cash flow and away from debt-heavy infrastructure bets, is the more defensible position going into 2027.
Frequently Asked Questions
Will the AI bubble burst in 2026?
Prediction markets currently price roughly a 17% chance of a full AI bubble burst by December 31, 2026, meaning traders see it as possible but not the base case. Central bank warnings from the BIS suggest the underlying risk — debt-financed capex and circular financing — is real regardless of the exact timing.
How is the AI bubble different from the dot-com crash?
Today's largest AI companies generate real profits, unlike most dot-com era firms. But 2025 brought a new risk the dot-com era didn't have at the same scale: rising debt financing of AI infrastructure, which the BIS says could turn a valuation correction into a credit event.
What is the BIS warning about AI and debt?
The Bank for International Settlements' 2026 annual report warns that AI infrastructure is increasingly financed with debt rather than cash flow, that credit spreads on that debt are widening, and that circular investment structures between chipmakers and AI labs could spread a shock through the financial system.
Is Michael Burry betting against AI stocks?
Yes. Burry has publicly positioned against an AI bubble he expects to burst, citing declining returns on invested capital for Big Tech's AI spending as evidence that valuations have detached from fundamentals.
What sectors would be hit hardest if the AI bubble bursts?
Cloud computing, semiconductor manufacturers, tech-hub commercial real estate, and the energy providers that built capacity for data centers face the most direct exposure. Because 2025-2026 AI capex leans more heavily on debt, corporate bond markets and the banks holding that debt are newly exposed as well.
Follow Peak of Trending for the next update on this story — including Q3 2026 earnings and capex guidance as they land.
Sources & References
- Bank for International Settlements — 2026 Annual Report warning on AI-linked debt and credit risk, via Seeking Alpha, 2026
- Gartner — Global AI spending forecast, via AL Capital Advisory, 2026
- Polymarket — "AI bubble burst by...?" prediction market data, Polymarket, 2026
- Forrester — 2026 AI market correction prediction, via TechTarget, 2026
- Moody's Analytics — Tech-sector corporate bond issuance data and Mark Zandi commentary, via ManageEngine Insights, 2026
- McKinsey & Company — Global workforce displacement projections, 2025
- Goldman Sachs — AI infrastructure investment tracking, Goldman Sachs, 2026
- GMO — OpenAI valuation and loss projections, GMO Research, 2026