Access to Power and the Debt Behind AI: A Box in the Bank of England's Financial Stability Report of December 2025

The Bank of England's Financial Policy Committee, whose primary responsibility is to contribute to the Bank's financial stability objective, published its Financial Stability Report in December 2025. Its summary says the role of debt financing in the AI sector is increasing quickly, as AI-focused firms "seek large-scale infrastructure investment." A box in the report, "Financial stability risks from the impact of AI development on financial markets," looks at that debt more closely. It uses "AI companies" broadly, for "a wide range of companies across multiple sectors," from technology to utilities and capital goods, "being at least partly dependent on AI progress for their current and expected future earnings." It names access to power among its sources of uncertainty. Parts of the box follow, not the whole.

From Cash to Debt

"AI infrastructure investment has mostly been financed by the cash flows of large, profitable technology companies and equity investments to date," the box says. "However, the financing of AI development is reaching an inflection point and while currently modest, debt financing is increasing quickly across multiple funding channels."

Part of the corporate debt issued by some AI companies in the second half of 2025, the box says, "has been raised through project-finance style off-balance sheet vehicles." Bank staff also estimate, using another firm's data, that new general lending to AI companies in what the box calls the "leveraged-finance and private market debt universe" increased "by a factor of three between 2023 and 2024 and then again further in 2025 to almost $100 billion year-to-date." That lending is made up of term loans, credit facilities and a limited share of bond issuance, the box says, and both banks and non-banks provide a significant share of it.

Where the risk ends up is hard to see. Some of the lending is later distributed or packaged into collateralized loan obligations, the box says, which spreads exposures "across a wide range of banks and non-bank investors." As a result, the box says, "tracing the final location of credit risk is challenging, complicating assessments of who ultimately bears losses if conditions deteriorate."

Debt Expected to Grow

The box expects more. "This debt financing is expected to increase significantly in the coming years," it says, and "a significant but uncertain share" of what AI infrastructure is expected to require "is likely to be financed by debt." Multiple firms noted the same expectation in recent market intelligence conducted by the Bank, it adds.

"The predominantly cash and equity-financed nature of AI development to date limits the likelihood of severe systemic risk implications of a change to AI-driven asset prices in the near term," the box says. Further on, it looks ahead. "If debt financing of AI development increases as projected this decade, the financial stability consequences to the UK economy of any AI-driven fall in asset prices could increase." Systemic banks could be exposed through their direct lending, the box says, and indirectly through non-bank financial institutions "such as private credit firms who are important providers of capital to AI projects."

The comfort in that judgment rests on how AI has been paid for so far, and the box itself expects that to change.

Access to Power

Among what it calls "other important sources of uncertainty," the box lists the degree of physical constraints to rapid AI infrastructure development, "most likely access to power." It also lists the depreciation lifecycle of AI chips, which it says comprise around half of AI data center costs, and the society-wide reaction to increasingly transformational AI. A later paragraph turns to power again. In the coming years, it says, the scale of AI demand for power could also increase wholesale electricity prices, with possible broader economic implications. The same paragraph ties power to credit. If access to power, for example because of delays in connecting new AI data centers to the electricity grid, "acts as a bottleneck to the operation of AI data centre projects, it can also weigh on their credit risk and be a trigger for re-evaluating future AI company/project earnings."

Read this way, a grid connection date belongs in the credit file as much as in the construction schedule.

The box puts it as a possibility, with "if" and "can," and gives no case. Who is holding that risk when a connection slips?