The Bank of England has done something no major central bank had done in a stability document before: it has treated the way the AI boom is financed as a financial-stability risk in its own right.
The July 2026 Financial Stability Report, published alongside the Financial Policy Committee’s 7 July record, spends real space on a question markets have mostly filed under “growth story.” How are AI companies paying for the data centres, chips and power contracts? Increasingly, the answer is debt.
The inflection point
The FPC dates the turn to 2025. That was the year, in the Bank’s telling, when the required investment of AI-focused companies exceeded their capacity to finance it from internal cashflows, pushing them toward external finance. In the first half of 2026 that shift accelerated, with AI firms turning to the financial system, particularly for debt financing, to fund infrastructure.
This is the part that matters for stability, and it is a genuine change. For most of the boom, hyperscaler capex was funded from operating cash and equity. Cash-rich balance sheets absorbed the spending. Once the spend outran the cash, the money had to come from somewhere, and it came from credit markets. The report points to public debt, private credit, leveraged finance and structured finance all expanding at what it calls an unprecedented pace.
That sentence is the story. It moves AI from an equity-market phenomenon, where losses land on shareholders who signed up for volatility, to a credit-market one, where losses land on lenders, funds and ultimately bank and private-credit balance sheets.
What the Bank said, and did not say
Read the report carefully and the framing is more careful than the headlines. The Bank is not declaring a live systemic threat. It is flagging a trajectory.
At the start of 2026, the FPC notes, the stock of outstanding debt from AI companies remained relatively modest, which helped to contain immediate risks to financial stability. The risk is forward-looking. The warning is conditional: if the scale of AI debt financing grows as expected over the coming years, an adverse shock to AI companies that results in losses or affects their ability to service debt could more materially affect global financing conditions.
So the honest summary is this. The Bank has named AI debt as a channel it now watches for systemic risk, while saying the channel is not yet large enough to be dangerous. That is a lower-key claim than “AI debt is a systemic risk today.” It is also, for a central bank, an unusually early flag. Stability reports tend to describe risks after they have grown. This one is describing one on the way up.
The assumptions the Bank is nervous about
The report is specific about what could go wrong, and the list is not financial in origin. It is operational. The Bank cites the size of future compute demand, the timely availability of power to data centres, and the depreciation rate of data centre facilities and AI chips as the assumptions that, if wrong, would hurt holders of AI debt.
That is a telling set of worries. None of them is about interest rates or credit spreads. They are about whether the physical AI buildout delivers the revenue the debt was raised against. If compute demand undershoots, if power arrives late, or if chips and buildings depreciate faster than the models assume, the cashflows that service the debt shrink. The lender, not the founder, wears it.
Why isolate the debt
The Bank already made AI a financial-stability workstream in the spring, when the FPC folded it into its 2026-2029 priorities. That earlier move was about governance, concentration and operational resilience: the risk that everyone runs the same model or the same cloud. The July report is narrower and, in a way, more concrete. It is about money owed.
Isolating the financing channel matters because it changes who is exposed and how a shock travels. A commoditisation event, where a cheaper model erases the pricing power of an expensive one, is an equity story if the buildout was equity-funded. It becomes a credit story, capable of tightening global financing conditions, once the same buildout sits on syndicated loans, private-credit funds and structured paper held across the system.
Implications
For lenders and private-credit managers, the report is a supervisory signal to price AI exposure as an operational bet, not a growth trade. For regulators elsewhere, the Bank has set a template: put the financing structure of the AI boom, not just its technology, inside the stability mandate.
And for the wider argument about an AI bubble, the Bank has quietly shifted the terms. The question is no longer only whether AI valuations are too high. It is whether the debt raised against them can be serviced if the physical buildout disappoints. Equity bubbles bruise investors. Credit ones move through the banking system. The Bank of England has now said, in writing, which one it is watching.
Discussion
Sign in to join the discussion.
No comments yet. Be the first to share your thoughts.