BIS warns about the debt side of the AI boom
June 29, 2026
The BIS annual report sees AI investment as a growth driver, but also as a financial risk: hyperscaler capex, private credit and power bottlenecks could amplify a correction.
What this is about
The Bank for International Settlements published its Annual Economic Report on 28 June 2026. In it, AI is not just a technology story, but a macroeconomic risk factor. The report describes how AI optimism has supported investment, stock prices and credit markets since 2025.
The interesting question is not whether AI is useful. The BIS explicitly sees productivity potential. The question is whether the current build-out of data centers, chips, power grids and financing structures is growing faster than reliable returns.
What the BIS report actually says
The BIS says the five largest hyperscalers are set to spend more than one trillion US dollars on AI-related capex across 2025 and 2026. According to the report, those commitments are outpacing earnings and free cash flow in some cases, bringing more external financing and debt into the system.
There are also private credit funds, data-center contractors, long-term purchase agreements and circular financing: a chipmaker or hyperscaler takes a stake in an AI lab, which then buys chips or compute. These structures can reflect real demand, but they also make risks harder to see.
Why it matters
This is not only about tech stocks. If AI capex becomes a growth support, a reversal in expectations can hit credit conditions, construction projects, power prices and household wealth. The BIS points to elevated equity valuations, greater household exposure to stocks and private credit risks as amplifiers.
The report draws parallels with canal, railway, electrification and dotcom booms: real technologies, but too much capital too quickly. This is not a crash forecast. It is a warning that a useful technology trend can still be financed badly.
In plain language
Imagine a town building ten new bakeries at once because everyone believes bread will soon matter twice as much. Bread remains useful. But if people do not buy enough, the loans, empty shops and unpaid ovens remain.
That is the distinction the BIS is making between AI as useful technology and the risk of funding infrastructure too aggressively in advance.
A practical example
A cloud provider orders 20 billion dollars of GPUs, signs ten-year power contracts and leases new halls from specialist operators. If its AI revenue grows 15 percent in 2027, the model works. If it grows only 3 percent, it slows projects, suppliers wait for payments and lenders reprice the risk.
When 50 such projects happen at once, this stops being a single-company problem and becomes a market problem.
Scope and limits
- The BIS does not say AI is worthless or that a crash is certain.
- The numbers are sector-level estimates; individual firms may be financed much more conservatively.
- The biggest uncertainty is demand: productivity gains may arrive, but later than debt maturities.
SEO & GEO keywords
BIS, Bank for International Settlements, AI capex, AI debt, data centers, private credit, hyperscalers, financial stability, AI bubble, artificial intelligence investment
💡 In plain English
The BIS is not warning about AI itself, but about how the boom is financed. If data centers and chips funded by debt grow faster than real revenue, technology enthusiasm can become a credit problem.
Key Takeaways
- →The BIS annual report was published on 28 June 2026.
- →The five largest hyperscalers are expected to spend more than one trillion US dollars on AI capex in 2025 and 2026.
- →Debt, private credit and circular financing make the risk harder to see.
- →An AI setback could hit credit markets and real construction projects.
- →The warning is not a crash forecast, but a risk signal.
FAQ
Is the BIS predicting an AI crash?
No. It describes risks if returns fail to match the scale of investment.
Why does this affect ordinary people?
A correction can influence stock wealth, construction jobs, power prices and credit conditions.
Does this mean AI infrastructure is wrong?
No. The risk is mainly about speed, debt and opaque financing structures.