BIS Flags AI Capex Financing Risk as Debt Replaces Cash Flow
The head of the Bank for International Settlements has placed AI capex financing risk at the centre of the global financial stability agenda, warning that the largest technology companies are spending faster than they earn and covering the gap with debt and private credit. Pablo Hernández de Cos, who leads the Basel-based institution, argued this week that the race between the biggest AI firms has become a credit story as much as a technology story, and that how the cycle ends will be determined by the funding structure as much as by the technology itself.
His comparison set was deliberate: the canal mania of the 1830s, Britain's railway boom of the 1840s, the electrification wave of the 1920s, and the dot-com surge of the late 1990s. Every one of those episodes rested on a genuine technological breakthrough, and every one still produced severe financial dislocation. The BIS reading of AI is that it belongs in that lineage.
What the BIS Means by AI Capex Financing Risk
The institution's annual economic report, published in June 2026, credits AI with supporting global growth while treating the investment surge as a stability question. According to the report, disappointment in returns could trigger a sudden pullback in financing, turning the capex boom into a protracted investment bust with knock-on effects for financial conditions more broadly.
Scale makes that channel consequential. The BIS estimates the five largest hyperscalers will spend more than $1 trillion on AI-related capital expenditure across 2025 and 2026 combined, a total that outpaces their earnings and free cash flow and has pushed several of them into debt markets to fund construction. The report adds that memory price inflation compounds the pressure, because each competitor is bidding to outbuild the others.
Hernández de Cos framed the exposure in terms of opacity and interconnection rather than raw size. Much of the funding for AI infrastructure sits in structures that are hard to see through, so supervisors cannot easily total up how much leverage has been created or who is holding it. Elevated valuations add a second channel, according to the BIS: when AI-linked shares are priced for sustained growth, an investment pullback hits equity and credit at the same time, and the two reinforce each other.
The Funding Stack Behind the Buildout
Three layers of capital sit beneath the data-centre buildout, and each behaves differently under stress.
- Corporate bonds. Issuance has run at record levels, and the BIS flags bond markets as particularly vulnerable if the AI cycle reverses.
- Private credit. Non-bank lenders have taken a growing share of AI-related borrowing, a channel with less disclosure and no public daily pricing.
- Off-balance-sheet vehicles. The BIS has warned that some financing structures built to support AI investment may mask leverage by shifting it off corporate balance sheets.
Leverage that has been moved out of sight does not stop existing; it stops being visible to the party being asked to underwrite it. Together, these layers form the AI capex financing risk the BIS is describing, and none of them is captured in a single company's reported debt figures.
Higher long-dated borrowing costs raise the benchmark against which AI-linked issuers are priced, a channel the BIS ties to record sovereign issuance competing for the same pool of capital.
Mapping the Three Pressure Points
The BIS does not present AI as a standalone risk. Its report groups the exposure into three linked pressure points that could crack the financial system together rather than separately.
| Pressure point | BIS assessment | How it transmits |
|---|---|---|
| AI capex bust | Spending outpacing earnings and free cash flow | Sudden financing pullback and a protracted investment bust |
| Circular financing | Opaque deals among a small set of linked firms | Hidden leverage and correlated counterparty exposure |
| Sovereign debt | Record issuance running alongside the AI buildout | Higher borrowing costs squeeze private borrowers |
Read together, the three entries describe a system in which the same handful of balance sheets appear on several sides of the same transactions. That is the interconnection Hernández de Cos singled out, and it is why the BIS risk case rests on opacity as much as on spending levels.
The sovereign channel closes a loop. Governments issuing record debt compete with AI borrowers for the same pool of long-dated capital, which lifts the yields that data-centre projects must clear before they are worth building, according to the BIS. Strain in public finances and strain in AI infrastructure financing feed each other instead of cancelling out.
Why the Historical Analogy Cuts Both Ways
Railways, electricity and canals all delivered on their promises. Investors in them frequently did not. The asymmetry matters: a technology can succeed while the financing that built it fails, because returns accrue over decades while debt matures in years.
The episodes also share an asset profile with the current buildout. Canals, rail lines and power grids were physical, long-lived assets funded by instruments that came due faster than the assets repaid their cost, and data centres financed through short-dated credit carry the same mismatch.
Applied to AI, the outcome hinges on a variable that is not technical. If commercialisation delivers at the scale vendors project, the trillion-dollar outlay becomes the largest infrastructure investment since electrification and the BIS warning reads as excessive caution. If it falls short, three pressures arrive at once: depreciation on specialised accelerators with contested economic lives, debt service on the borrowings that funded them, and stress across private credit funds holding the junior claims.
The BIS treats the downside as more than a market correction. Its framing puts AI data-centre investment alongside sovereign debt as a question of global financial stability rather than a sector-specific industrial cycle.
Who Carries the Residual Risk
The question that matters for boards, lenders and their supervisors is who ends up holding the loss. Under the current structure, the answer is distributed and partly invisible.
Hyperscalers retain equity risk on their own balance sheets, but they have shifted a growing share of the capital cost onto bondholders, private credit funds and special-purpose vehicles. Those counterparties hold claims on assets whose useful economic life is contested and whose resale value depends on a secondary market that has not been tested at scale.
That arrangement works well for the hyperscalers while financing remains available and spreads stay tight. It concentrates risk in lenders who cannot easily establish how much of the same exposure they hold indirectly through other funds, which is the practical meaning of an opaque and interconnected funding chain.
Financing mix also becomes a competitive variable. A hyperscaler funding construction from operating cash flow keeps the option to build through a credit squeeze, while a rival leaning on private credit and project vehicles faces a higher cost of capital first. The pace of the AI buildout is now set partly by credit markets rather than by chip supply or model quality alone.
The signals to watch therefore sit in funding markets rather than in construction data. A tightening of private credit terms, wider spreads on AI-linked issuers, a slowdown in circular deal flow, or further increases in long-dated Treasury yields would each show up before any visible decline in data-centre building. Supervisors have started treating those indicators as early warnings because they identify where the residual risk has accumulated.
Why this matters
The BIS has reframed the AI buildout from an industrial race into a credit cycle with a long tail, and the AI capex financing risk it describes sits with lenders and investors rather than with the companies ordering the hardware. Firms planning multi-year compute commitments should test their assumptions against a scenario in which financing costs rise and counterparties reprice, alongside the scenario in which demand keeps compounding. The technology can be transformative and the financing behind it fragile at the same time, and the second condition decides how much of the first survives.
Related Articles
- AI Infrastructure Spending Risk Could Trigger Global Financial Crash, BIS Warns
- Hyperscaler Bond Issuance Surges 18-Fold as AI's $220 Billion Debt Bill Tests Investor Patience
- 2026 hyperscaler capex hits $697B as AI race turns to debt
✔Human Verified
Researched and cross-referenced against primary sources by the Bytevyte editorial team. This article was generated with the assistance of artificial intelligence and reviewed by the Bytevyte editorial team.