AI debt risks center on the massive, increasingly leveraged and opaque financing of the AI infrastructure boom (primarily data centers, GPUs, power and related equipment). As of late 2026, this has shifted from largely cash-flow-funded expansion by hyperscalers to heavy reliance on debt and complex off-balance-sheet structures. Regulators (Fed, ECB) and analysts (Goldman Sachs, The Economist, IMF) highlight elevated risks of a correction if AI revenues disappoint with potential spillovers to credit markets, banks and the broader economy—though a full systemic crisis is viewed as less likely than a sharp equity/credit repricing.
Scale of the Debt Build-Up
- On-balance-sheet debt : The seven largest AI-related firms (Amazon, Alphabet, Meta, Microsoft, Nvidia, Oracle, Broadcom, and similar) are expected to exceed $800 billion in on-balance-sheet debt by end-2026. Hyperscalers alone issued roughly $296 billion in 2026 (above earlier forecasts of ~$250 billion) with Goldman Sachs projecting $420–440 billion in issuance for 2027.
- Off-balance-sheet and contingent commitments : Far larger—Fed estimates >$3 trillion in maximum notional value (leases not yet started >$1 trillion + take-or-pay contracts and residual-value guarantees >$2 trillion). ECB estimates exceed $4 trillion for a similar group of firms. Goldman Sachs puts key hyperscalers’ lease + purchase commitments around $3.2 trillion.
- Broader AI-related debt : Advanced-computing debt issuance exceeded $680 billion by October 2026 (vs. $130 billion in all of 2023), accounting for ~31% of nonfinancial investment-grade corporate issuance. Total AI-linked debt exposure is estimated in the high hundreds of billions, with private credit and project finance adding more. Goldman projects significant project finance growth into 2027 (hundreds of billions for shells and chips).
Hyperscalers’ free cash flow turned negative for most in 2026 as AI capex outstripped operating cash flow. Capex for the top players is forecast at $1.2–1.4 trillion in 2027.
Key Risk Factors
1. Opacity and complexity : Much financing uses special-purpose vehicles (SPVs), long-term leases, residual-value guarantees (especially on GPUs) and take-or-pay contracts. These keep obligations off main balance sheets initially. Risk location is hard to track—held by high-yield funds, insurers, private credit, BDCs, and banks (largest U.S. bank holding companies have ~$1 trillion exposure to tech, data centers, SPEs, private credit and utilities).
2. Circular and interconnected financing : Nvidia (and to a lesser extent others like Alphabet/Broadcom) provides equity stakes, capacity guarantees, and residual-value backstops (Nvidia’s contingent liabilities estimated in the hundreds of billions, including large deals tied to OpenAI and neoclouds). Customers buy chips/capacity with financing supported by the supplier, creating feedback loops. If demand softens, equity losses, debt impairments and revenue shortfalls can hit simultaneously.
3. Asset-liability mismatch : AI chips/servers have short economic lives (often 1–3 years estimated by market intelligence) vs. longer debt/lease maturities. Accounting lives have been extended (e.g., to 6 years), but rapid technological obsolescence or overcapacity could impair collateral values.
4. Revenue realization lag and demand risk : Huge upfront capex requires substantial future AI revenues (e.g., estimates of hundreds of billions annually needed for reasonable returns). Enterprise AI spending has shown signs of plateauing in some data; slower adoption, open-source competition, energy/grid constraints or regulatory pushback (e.g., data-center restrictions) could leave capacity underutilized.
5. Weaker credits and neoclouds : Pure-play data-center operators and neoclouds (e.g., CoreWeave) carry higher leverage and face rising yields/spreads. Some bond yields have reached distress levels; CDS imply elevated default probabilities. Banks and investors have grown more cautious post-market volatility.
6. Interest rates, refinancing, and credit conditions : Higher rates squeeze coverage. Spreads on AI-related bonds have widened relative to peers. A broader credit tightening or recession could amplify problems.
Potential Transmission and Severity
- Direct impacts : Equity correction in AI/tech stocks (already highly concentrated), credit losses for private credit/insurers/banks holding the riskier tranches/project delays/cancellations.
- Broader effects : Wealth effects on consumers, reduced investment, and possible spillover to utilities/power markets. ECB has drawn parallels to subprime (opacity, off-balance-sheet leverage, circularity) more than pure equity-funded tech booms like the late-1990s dot-com era (which caused a shallower recession partly because equity absorbed more losses).
- Mitigants : Hyperscalers retain strong overall credit profiles and diversified cash flows from non-AI businesses. Many obligations are contingent or supported by investment-grade parents. Physical assets (buildings, some power infrastructure) retain value even if GPUs depreciate. Contagion to the core banking system appears more limited than in 2008 so far, though private credit opacity is a concern. The Economist and others note a manageable equity crash is more plausible than a deep financial crisis.
Outlook and Monitoring Points
Risks are rising into 2027 as debt issuance is expected to accelerate further and more commitments convert to funded liabilities. Key watch items include :
- Actual vs. expected AI revenue growth and utilization rates.
- Further widening of credit spreads or rating actions.
- Capex guidance and free-cash-flow trajectories from hyperscalers.
- Regulatory scrutiny (Fed Financial Stability Report already flags this, ECB similarly concerned).
- GPU secondary-market prices and residual-value realizations.
In summary, AI debt has scaled rapidly into the trillions when including contingent exposures, introducing classic leverage, maturity, and opacity risks on top of technology execution uncertainty. While the strongest players can absorb setbacks, a significant disappointment in AI monetization could trigger a meaningful market correction with credit-market amplification—exactly the dynamic The Economist highlighted in its 2027 outlook. The situation remains fluid and data-dependent.