Bank of England Warns: AI Infrastructure Is Entering the Debt Market
The next phase of the AI boom needs not only chips and power, but also ever more debt. The Bank of England says financing for data centers and relate…
On July 7, the Bank of England wrote something that had previously been seen more as a technology-industry issue into its Financial Stability Report: the expansion of AI infrastructure is increasingly reliant on debt financing. Data centers, chip purchases and supporting energy projects are still being built at pace, but the sources funding them are no longer limited to companies’ own cash flow; they also include the public bond market, bank loans, private credit, leveraged finance and asset-backed financing tools.
What happened
The Bank of England said AI-related companies significantly increased external financing in the first half of 2026, especially debt financing. The report said financing channels for AI investment are becoming more diversified, and risks are therefore spreading across a broader credit ecosystem.
One key change is that large cloud providers and data center projects have begun using off-balance-sheet financing, special purpose vehicles, data center asset-backed structures and other customized arrangements. These structures can place construction costs and part of the debt at the project or partnership entity level rather than showing everything on the parent company’s balance sheet. The Bank of England believes this makes it harder for financial institutions to identify where the ultimate risk lies.
The report also noted that the share of AI investment financing provided by private credit has risen. Estimates cited by the Bank of England show private credit’s share of AI investment financing increasing from 9% in 2024 to 34% in 2025. At the same time, the central bank stressed that as of early 2026, the stock of AI companies’ debt remained relatively limited compared with the overall financial system, so direct systemic risk is still contained for now.
Why it matters
This is not about judging whether AI demand exists, but about asking: if future profitability materializes more slowly than capital spending, who will bear the credit risk during the buildout phase.
AI infrastructure has the characteristics of high fixed costs, long construction cycles and heavy supply-chain dependence. Projects must first spend heavily on chips, build data halls, secure power and network capacity, and only then generate cash flow through cloud services, model usage or enterprise contracts. As long as financing conditions remain loose, this cycle can continue; but if interest rates rise, leases are delayed, customer demand falls short of expectations, or future AI revenue cannot cover refinancing costs, risk could transmit through bonds, private credit, bank facilities and asset securitization structures.
The Bank of England specifically noted that the complexity and lack of transparency in AI-related debt structures could amplify market shocks. Reuters cited the report as saying that if investors’ bets on AI success are re-priced, share price declines could reinforce each other with high concentrations, correlated trades and leveraged positions. In other words, the key variable in the AI trade is expanding from “can the model get stronger?” to “can cash flow support the capital structure?”
This also explains why AI infrastructure is becoming a market-structure issue. Previously, investors mainly participated in AI growth through tech stocks and chip stocks; now, more capital may enter the same industrial chain through corporate bonds, private funds, project finance and asset-backed securities. The sources of return are more diversified, but the risks are also harder to observe fully from public market prices.
What to watch
First, AI companies’ refinancing needs over the next two years, and whether debt maturities match the cash flow cycles of data center projects. Second, whether private credit funds and banks are indirectly exposed to the same set of projects through loans, committed lines or fund interests. Third, whether leases, guarantees and customer commitments in off-balance-sheet structures are sufficient to support creditors’ claims. Fourth, when large cloud providers cut capital expenditure or delay projects, whether knock-on effects emerge among chips, data center operators and energy infrastructure.
The Financial Stability Board also warned in June that private credit has been growing rapidly, while some markets have yet to undergo stress testing under a prolonged economic downturn. The Bank of England’s warning should therefore not be read as saying the AI bubble has already burst, but more as a risk map: the next wave of AI expansion may need the credit market to complete it; and the credit market will ultimately ask it to prove that future revenue is sufficient to repay today’s infrastructure bill.
Sources
Information only. No investment, legal, tax, or financial advice.