Banks are now trying to offload the AI data center debt they just underwrote

What does it mean that banks are trying to offload AI data centre debt?

Major banks are reportedly trying to move part of their AI data centre lending to other investors. The Financial Times, reported via Silicon UK on 5 May 2026, said JPMorgan, Morgan Stanley, SMBC and MUFG are nearing internal exposure limits on single borrowers and sectors. This is a capacity problem, not a sign of collapse.

The detail matters. JPMorgan, MUFG and other lenders spent more than six months distributing $38 billion (about £28bn) of construction debt for a Texas and Wisconsin project leased to Oracle, and some banks sold loans at a discount to non-bank lenders. Morgan Stanley has also held early talks about a significant risk transfer, with no guarantee of a deal.

How do significant risk transfers work?

In a significant risk transfer, a bank pays investors to absorb losses on part of a loan pool, typically around 5% to 15%, while keeping some exposure itself. The loans stay on the bank's books, but part of the risk moves elsewhere, which frees up room to lend more.

Critics argue the debt is moved rather than removed, often into private credit funds that face less regulatory scrutiny. Reports note there is no evidence of a wave of defaults. The pressure comes from exposure limits and sheer issuance volume. Oracle and CoreWeave have borrowed hundreds of billions of dollars, and Man Group's Matthew Moniot called the scale "out of scale to anything we've thought about, ever."

Should small businesses be worried about this?

Not about a bank failure. Nothing reported suggests that. The sensible concern is dependence: the AI tools you build your business on are priced and funded by an industry that is borrowing at unprecedented scale. If that funding tightens, tool pricing, limits and terms can change.

Local opposition is another cost risk. Maine recently passed a statewide moratorium on new data centres, and public resistance can slow or raise the cost of projects. None of this means AI tools will vanish. It means you should not assume today's prices and terms are permanent.

What should founders do about it?

Treat AI as a replaceable layer, not a foundation. Use it to speed up work, but keep your customers, bookings and content in systems you control. If one tool changes its price or disappears, you swap it out without losing your business.

1. List your AI dependencies. Note every tool that would hurt if its price doubled or access was cut.
2. Keep exports current. Customer lists, prompts, templates and content should live in files you own.
3. Avoid long prepaid lock-ins on tools that have no clear alternative.
4. Budget a buffer. Assume software costs may rise and plan for it.

Is this the same mistake as relying on rented platforms?

Yes, and it is the strongest argument for owning your infrastructure. Whether the risk is a commission, a platform policy or a funding squeeze upstream, the pattern is the same: you rent the customer relationship, so you do not control the terms.

We have seen the cost of this directly. A nail salon was paying Fresha about £1,800 a month in commission. Moving to its own booking system at £35 a month saved roughly £21,000 a year. A plumber paying £280 a month to Yell for 3 leads now gets 18 to 22 enquiries a month from a Google Business Profile and a £120 Google Ads campaign. Owned channels simply give you options.

Braynex Services' view: should you slow down on AI?

No. Our view at Braynex Services is that the right response to a financing story is not to pull back from AI, but to adopt it on your own terms. Use it where it saves real time, such as answering enquiries, following up missed calls and drafting content, and build it on top of systems you own.

The businesses that suffer when a technology cycle wobbles are those whose entire operation sits on someone else's platform. The ones that cope have their own website, their own booking and their own customer data. That is also what gets you recommended by assistants like ChatGPT, Gemini and Google AI Overviews, because those tools need a real, well-structured source to cite.

What is the next step?

Pick one dependency from your list this week and reduce it: export your customer data, or move one rented service onto something you own. If you would like help spotting where you are exposed, book a free audit at braynexservices.com and we will show you where your business is renting what it should own.