Who governs AI infrastructure when its financing fails?
As AI becomes embedded in essential services, its providers’ financial stability matters beyond the technology sector. In England, AI tools that help radiologists review chest X-rays are available in around half of National Health Service (NHS) trusts. In the latest Bank of England and Financial Conduct Authority survey, announced in October 2026, 86% of UK financial firms reported using AI, including for fraud detection. If a critical provider fails, diagnostic reviews could be delayed or fraud monitoring disrupted where alternative arrangements cannot be made quickly. Service contracts may set availability commitments and switching rights, but do not necessarily secure the funding, technical access, and operational support needed to maintain or transfer a service during financial distress.
The way AI infrastructure is financed makes that problem harder to see. The Bank for International Settlements (Switzerland) describes a common arrangement in which a separate company, often called a special-purpose vehicle, owns or develops a data centre and borrows to fund it. A large technology company leases the facility or commits to purchasing computing capacity, and those payments repay the debt. Much of the borrowing therefore sits outside the technology company’s balance sheet. The Bank of England warns that data centres unable to support newer AI chips may become outdated sooner than expected, reducing their value and the cash available to repay debt.
In the Internet Policy Review, Luitse (2024) has shown how cloud platforms exercise infrastructural power in AI. Financing adds another layer. The organisation using a service often contracts with an AI supplier several steps removed from the data-centre owner and its lenders. If any link in that chain defaults, those lenders can shape whether the service continues or moves.
When continuity sits outside the service contract
The OECD’s 2026 Global Debt Report examines a financing involving Meta and Blue Owl Capital for a large AI data-centre campus in Louisiana. A separate company, Beignet Investor LLC, issued US$27.3 billion in debt maturing in 2049. Meta agreed to lease the facilities and, subject to conditions, make a capped payment if the leases end and the property’s value falls below an agreed threshold. These terms address how investors are repaid if the leases end. They do not establish whether a service relying on that infrastructure could continue or move in financial distress.
What distress looks like in practice
Builder.ai, a software developer, entered insolvency proceedings in 2025 after its lender seized most of its cash. Published accounts give little detail about what happened to customers’ unfinished projects. Two earlier infrastructure failures are better documented.
UKCloud was a British cloud provider used by public bodies. In October 2022, a court ordered its compulsory liquidation, closing the company and selling its assets to pay creditors. To help maintain public services, the Cabinet Office provided £20 million in working capital and agreed to cover certain costs and liabilities of the Official Receiver, the government official overseeing the liquidation. By February 2023, UKCloud’s services had ceased and customers had moved to other suppliers, reportedly without unexpected disruption to public services. Public money had kept services running while customers left.
Cyxtera, a US data-centre operator, took a different route. It entered Chapter 11 proceedings in June 2023, a court-supervised process that generally lets a company keep operating while restructuring its debts. Some existing lenders committed up to US$200 million, subject to court approval, to fund operations during the restructuring, and Cyxtera reported that its platform continued operating “normally and without interruption”.
The continuity-funding gap
Both cases show that continuity during financial distress can depend on funding. This op-ed uses the term continuity-funding gap for the shortfall between the money needed to keep, move, or safely close a service and the funds that can lawfully be used in time. Funding that arrives after a service’s tolerable period of disruption has passed may be too late to prevent harm.
Even with funding, a transfer may need data, encryption keys, licences, staff, and replacement capacity. For proprietary AI services, exporting data may not preserve the model’s function. A replacement may also need testing and validation before the service resumes.
Existing rules address some of these risks. The EU Data Act requires providers of cloud and other data processing services to remove obstacles to switching and help customers move to another provider or their own systems (arts. 23, 25). For financial firms, the EU’s Digital Operational Resilience Act requires tested exit plans for technology services supporting critical or important functions (arts. 28(8), 30(3)(f)). In the UK, regulators began overseeing four Treasury-designated critical technology providers in July 2026. Australia’s CPS 230 requires regulated institutions to set disruption limits for critical operations and to be able to exit material service arrangements in an orderly way. The Financial Stability Board’s 2023 toolkit treats a supplier’s deteriorating financial condition as a reason to plan for exit. Apart from the Data Act, these measures focus on financial services. They improve preparedness but do not establish who can release the funding or access needed to maintain or move a service during insolvency.
Building continuity before distress
For any organisation relying on an AI service, a practical first step is a counterparty continuity schedule: a short record of the organisations, assets, contracts, accounts, and data the service depends on, including lenders and others outside the customer’s control.
Next, test the funding route: who would pay to keep, move, or close the service if the provider defaults? Money held in escrow or an on-demand bank guarantee usually remains available, subject to its release terms. A credit line to the provider usually does not, because lenders can stop lending after a default. Where no such money is secure, arranging a back-up provider early may be the better control.
Buyers can require insolvency continuity plans, updated when suppliers experience financial distress, as UK central government guidance already expects for critical service contracts. Regulators and major buyers could extend this approach to important AI services by seeking evidence of who controls the resources needed for continuity and how the associated costs would be funded.
For an essential service, “exit” means moving it to another provider, bringing it in-house or closing it safely without stranding users. Planning exit before distress would let hospitals, banks, and public agencies keep services running through a supplier’s failure, and reward providers with credible continuity plans over emergency rescues like UKCloud’s. Settling continuity costs in contracts before failure would make the AI infrastructure the public depends on more durable and financially sustainable.