Bank Of England Warns Of AI Debt Risks
As the Bank of England warns that growing borrowing is increasing the financial risks surrounding AI, here we look at how the technology’s expansion is being funded, why expected returns could disappoint and what businesses should consider before becoming dependent on its providers.
What Has The Bank Warned About?
The Bank’s Financial Policy Committee, which monitors threats to financial stability, discussed AI financing at its meeting on 25 September. Its published record warns that increasing borrowing and complicated financial relationships could spread losses if the technology delivers less than investors expect.
The concern extends beyond falling technology shares. For example, lenders and other investors are increasingly helping finance the infrastructure behind AI, giving them a stake in whether anticipated demand becomes profitable business. The committee warns that these arrangements could amplify losses if expectations are not met.
Why Does Borrowing Change The Risk?
Building AI services requires substantial spending before the resulting income is certain. For example, data centres need buildings, electricity supplies, cooling and computing equipment, while developers must fund research and the ongoing cost of answering customers’ requests.
When investors buy shares, they accept that their returns depend on the company’s performance. Borrowing creates repayment obligations that remain even if sales grow slowly, customers negotiate lower prices or equipment becomes outdated sooner than expected.
A company can therefore develop useful technology while struggling financially. Strong demand doesn’t automatically produce sufficient profit, particularly when serving more customers also increases operating costs and competition limits what the supplier can charge.
How Large Is The Financing?
The scale of the borrowing is already pretty substantial, with the Bank citing Morgan Stanley’s estimate that AI-related businesses worldwide had raised around US$450 billion through debt by early September, more than double the total for 2025. That borrowing could grow considerably, with JPMorgan forecasting that debt will fund approximately US$4.1 trillion of investment in AI infrastructure and equipment between 2026 and 2030.
It should be noted here, though, that both figures cover the wider AI sector, rather than a single company, and the longer-term estimate depends on how investment develops over the next few years.
What Are Circular Financing Arrangements?
The financial picture becomes a bit more complicated when a company invests in an AI developer that then uses the money to buy cloud services or computing equipment from that same investor. These arrangements aren’t necessarily improper, but they can make it harder to tell how much of the reported growth reflects demand from independent customers and how much relies on investment flowing back to the companies providing it.
For example, an infrastructure supplier may benefit from a developer’s spending while also depending on that developer becoming successful enough to keep paying. If funding dries up, both sides could face pressure, with consequences for other companies relying on them.
What Does Anthropic’s Reported Spending Show?
Details reported from Anthropic’s IPO prospectus illustrate how rapidly an AI company’s costs can grow even as sales increase. The document sets out its plans for an initial public offering, through which it would sell shares to public investors.
The reported figures show that Anthropic generated nearly US$4.6 billion in revenue in 2025 but recorded an operating loss exceeding US$8 billion. Its overall net loss was much larger at around US$42 billion, although roughly US$34 billion of that came from accounting charges linked to financing arrangements, rather than money spent running the business.
The prospectus also reportedly sets out US$518 billion in future cloud and infrastructure commitments, showing the scale of spending planned over the coming years rather than a bill due in a single year.
Could Competition Make Repayment Harder?
As competing AI models improve, customers may be able to get the results they need at lower prices, putting pressure on the profits providers are relying on to recover their investment. That competition includes open-weight models, which make the settings learned during training available for others to use or adapt, allowing businesses to run them themselves or choose a company to host them.
However, running these models still involves costs for hardware, technical support and security, so a cheaper alternative is not automatically cheaper overall. Businesses need to compare the full cost of getting a task done reliably, including connecting the model to existing systems, supervising its work and correcting mistakes, before deciding whether a lower price represents a worthwhile saving.
What Does This Mean For Your Business?
For businesses buying AI services, the warning highlights how supplier resilience should be part of the purchasing decision. A convincing demonstration should be followed by questions about service continuity, pricing commitments and what happens if the provider changes ownership or withdraws a product. Where an intermediary supplies the service, businesses should also understand which underlying model and cloud providers it depends on.
Contracts and implementation choices should preserve a practical way to move elsewhere. Keeping accessible copies of business information, documenting integrations and testing alternatives can reduce disruption if prices rise or service quality falls. That doesn’t require avoiding newer suppliers, but the importance of the task should determine how much dependence is acceptable.
Businesses should really base their own AI spending on what the technology can do for them, using a limited trial to test whether improvements in speed and accuracy justify the cost of subscriptions, setup and staff time spent checking the work. Even if some AI investors lose money, customers could still benefit from useful services, provided the savings or improvements hold up in everyday use and do not depend on prices staying artificially low.
