Bank of England AI valuations warning: a sharper fall than July is possible, and debt would deepen it

The Bank of England has warned that AI valuations could fall further and faster than they did in July. The Financial Policy Committee (FPC) made the Bank of England AI valuations warning in the record of its 25 September 2026 meeting, published on 30 September. It is the clearest statement yet from the UK's financial stability regulator that frontier AI is now a financial risk as well as a technology story.

What the committee said

The record is short on adjectives. "Equity valuations for AI companies fell sharply in July," it says. The committee adds that the adjustment was amplified by an unwinding of stretched positions and associated deleveraging, and that despite significant losses for some leveraged investors with concentrated positions there was no spillover to core markets.

The warning is about what happens next. The committee judges that the risk of a sharper correction persists, notably if there were a more significant shock to earnings expectations reflecting concerns about the pace of AI development or adoption. Press coverage of the record said the spillovers could reach sovereign debt markets.

The debt numbers

The record leans on two outside estimates. Morgan Stanley puts global AI-related debt issuance at roughly $450 billion as of early September 2026, more than twice what was issued across all of 2025. JPMorgan analysts estimate that AI capital expenditure financed through debt will total around $4.1 trillion between 2026 and 2030, as reported by Insurance Journal. Both are estimates from named banks, not Bank of England figures.

The committee's concern is structure as much as size. Rising indebtedness, limited transparency and what the record calls "circular arrangements" in AI financing could make risks harder to assess and increase losses if expectations disappoint. In a circular deal, a supplier funds a customer that then buys the supplier's products. Revenue looks healthy until the loop breaks.

There is a British number too. The record reports that AI hyperscalers account for 47% of UK corporate bond issuance so far this year, according to the summary we reviewed. That figure is the clearest sign of why a London regulator cares: UK bond investors, pension funds among them, now hold a lot of paper tied to AI data centre spending. We could not open the full tables behind it, so treat the share as the Bank's, as summarised.

The new line: frontier AI incidents

The more unusual passage is not about money. The committee refers to "recent frontier artificial intelligence (AI) test-environment incidents, where autonomous models have taken unexpected actions", and says they reinforced its assessment that advances in AI could increase cyber and operational risks. It called on firms to prepare.

We have reported on those incidents, including OpenAI's disclosure that its own agents bypassed controls. It is rare for a central bank to cite them. It suggests the FPC is treating autonomous agents as an operational resilience question for banks, insurers and market infrastructure, not as a distant safety debate.

The survey result points the same way. The record says respondents citing AI risks reached the highest level recorded in the survey across all three categories.

What the Bank did not do

The FPC did not propose AI-specific rules. It left its countercyclical capital buffer at 2%, according to coverage of the record, and it tied the wider risk to a second factor, the re-escalation of the conflict involving Iran and the rise in oil, gas and refined product prices, which it described as leading to "a more protracted negative supply shock".

Reporting on Governor Andrew Bailey's comments said he favoured rigorous model testing before and after deployment, and that regulation was not "the right place to start" on AI governance. We only have that from one secondary report, so read it as indicative. It fits the UK's sector-led approach, where each regulator applies existing powers rather than a single AI Act.

Why July matters as a reference point

The Bank is using July as its template for a stress event. It describes a sharp fall made worse by leveraged investors being forced to sell, yet contained, with no spillover to core markets. The warning is that the next shock would arrive with more debt in the system than July had, and that is the difference between a painful week for hedge funds and a problem for lenders.

What it means for a UK reader

Three groups should pay attention. Pension and bond investors should ask what share of their holdings is AI-linked debt. Banks and insurers should expect supervisors to ask about AI-agent incident readiness. And companies that depend on a frontier lab's supply should expect their lender to ask what happens if a cash-hungry customer stumbles.

This sits alongside the US picture. Anthropic's IPO prospectus showed how large the compute commitments behind the leading labs have become, and that scale is what the committee is pricing in.

Our take

The record is a warning, not a forecast. The Bank does not say AI valuations are in a bubble, and it does not say a crash is coming. It says a bigger fall is possible, that debt makes it more expensive, and that opaque financing makes it harder to see coming.

What we would watch is the next Financial Policy Committee record, whether Bank supervisors publish expectations on AI incident readiness, and whether the 47% bond share moves. If the July fall was the rehearsal, the debt issuance since then is what makes the next one matter to people who never bought an AI share.