Bank of England chief warns AI bubble burst would be a “disaster” for the world economy
Andrew Bailey has used his position as chair of the G20’s Financial Stability Board to issue one of the starkest warnings yet from a central banker on the risks building up around artificial intelligence investment. In a letter sent to G20 finance ministers gathered in Asheville, North Carolina, this week, the Bank of England governor said markets remain exposed to a mix of stretched valuations, rising debt and growing concentration in the AI sector, and that a “future market correction” could ripple across borders.
What sets Bailey’s intervention apart from earlier central bank commentary is where he pointed the finger. Rather than treating this purely as a stock market story, he flagged the deepening financial ties between AI developers and the hyperscale technology firms building the infrastructure that powers them, an unusually direct nod to the data centre sector.
This is no longer a warning confined to Wall Street analysts fretting over Nvidia’s share price. It’s a regulator naming infrastructure financing as a transmission channel for systemic risk.
The financing web tying AI labs to the data centre build-out
The structures Bailey is alluding to have been building for months. Multi-billion pound commitments between chipmakers, cloud providers and AI labs, of the kind Capacity has tracked in its coverage of the “AI house of cards” financing question, increasingly rely on circular arrangements: one company’s capital expenditure becomes another’s revenue, often financed through debt rather than cash on hand.
The Bank of England’s Financial Policy Committee has been flagging the resulting exposure for over a year, warning in its minutes that valuations for AI-focused technology firms “appear stretched” and drawing direct comparisons with the dotcom era. Bailey’s letter reiterates that view at G20 level, adding that these vulnerabilities interact with strains elsewhere, including sovereign debt markets and geopolitical shocks.
For data centre operators, the practical question is what happens to project financing if that correction arrives. A share price shock at a hyperscaler doesn’t stay contained to equity markets when that same company is guaranteeing offtake agreements, backing special purpose vehicles, or underwriting the debt behind a campus still under construction. Capacity’s own reporting on Microsoft’s lease cancellations across the US, UK and Europe already showed how quickly a hyperscaler can retreat from committed capacity when its own capital allocation comes under scrutiny, long before any formal “bubble” narrative took hold.
Five companies, one sector’s fortunes
Bailey’s letter lands alongside a striking data point from the Bank’s own analysis: the five largest US technology companies now account for roughly 30% of the S&P 500’s total value, the highest concentration in half a century. Nvidia alone sits close to $4.5 trillion, with Microsoft not far behind.
That concentration is precisely why the data centre sector has become so sensitive to sentiment in AI markets. Demand forecasting for new capacity has, in effect, been outsourced to the capital expenditure plans of a handful of companies. When TD Cowen analysts described lease cancellations as pointing to “data centre oversupply relative to its current demand forecast”, as covered in Capacity’s report on Microsoft’s retreating AI footprint, they were describing exactly the kind of correction Bailey now warns could spread internationally.
Not everyone in the industry sees it that way. Alibaba chair Joe Tsai told an investment summit last year that he was watching the numbers with concern, saying: “I start to see the beginning of some kind of bubble,” as Capacity reported at the time. Others remain firmly unconvinced. Sparkle CEO Enrico Starace, speaking at Metro Connect earlier this year, pushed back on the framing entirely, telling delegates: “This is not a bubble, it’s fed by real demand,” a comment Capacity covered in its analysis of capital keeping pace with data centre demand.
That gap between regulator caution and operator confidence is arguably the more interesting story than the warning itself. Central banks are assessing systemic exposure across the whole financial system. Data centre operators and hyperscalers are assessing whether the physical demand for compute justifies the capital being committed. Those are related questions, but they are not the same question, and the answer each side gives depends heavily on which risks they are mandated to worry about.
What this means for infrastructure decision-makers
Bailey’s letter also touched on a second, less discussed risk: the growing capability of frontier AI systems to be used in cyber attacks, which he described as the “most immediate concern” for financial stability. For an industry already contending with heightened scrutiny of data centre security and resilience, that’s a second front worth watching alongside the financing question.
None of this means the data centre build-out grinds to a halt. Power procurement, land acquisition and construction pipelines already underway are not easily reversed, and underlying demand for compute, whatever its ultimate scale, has not disappeared. But Bailey’s letter is a signal that the industry’s financing assumptions are now being scrutinised at the highest levels of global economic policymaking, not just by equity analysts.
RELATED STORIES
What is a neocloud, and why does it matter for digital infrastructure?
What is sovereign AI infrastructure, and why does it matter for telecoms?
What is direct-to-device satellite connectivity?
What is the Ratepayer Protection Pledge, and why does it matter for data centres?






