Insurers Are Now the Risk-Sink for AI Compute: Barclays Is Buying Credit Protection on More Than $10 Billion of Data-Center Debt

Barclays is doing something banks rarely do with a loan they still want: keeping the facility and the fee while pushing the default risk somewhere else. Bloomberg reported on Oct 28, 2026 that the bank is looking to buy default protection on more than $10 billion of AI data-center debt, and that it has asked insurers to cover more than 50% of the credit risk on some of those facilities. Barclays is selling the default leg of AI infrastructure loans without selling the loans themselves, which means it can keep originating into the build-out without funding the tail from its own capital.

None of that is stressed. Barclays earned £8.06 billion over the trailing twelve months at a 10.8% return on equity while managing toward the top of its 13–14% CET1 range and absorbing risk-weighted asset headwinds into 2027. A premium paid to an insurer is a rounding error against £8.06 billion of earnings, which is exactly why the bank would rather insure the tail than retreat from AI lending.

The wider context is a balance sheet that has been leaning into non-bank lending: £66 billion of structured financing to non-bank financial institutions at the end of March, of which £15 billion is private credit carrying roughly £300 million of mark-to-market. Insurer protection is how Barclays keeps that book growing while lowering its own capital intensity, and it tells you the bank expects to do more of it, not less.

Barclays' move looks novel and is not. Insurance capital has been sitting inside AI data-center loans for months, most visibly as the anchor on CoreWeave's largest facilities. Blackstone's credit and insurance unit underwrote the $8.5 billion loan that JPMorgan Chase and Goldman Sachs arranged in March 2026 — the first investment-grade-rated GPU-backed facility of its kind, with a syndicate that already included Barclays.

The link to Barclays is direct, but not confirmed as the same trade. Barclays has been underwriting AI data-center exposure for two years, and the borrowers it is now hedging are not disclosed, so CoreWeave cannot be named as the target. CoreWeave is simply the clearest documented case of a bank holding live AI credit that insurers already helped finance — and the size of the request, billions rather than millions, says Barclays is hedging a book, not one facility.

The gap between the backlog and the bonds is the trade Barclays is making. Insurers are being asked to write protection against a credit that repriced roughly 1,000 basis points in a matter of weeks, and the premium they charge will be set by the loss assumption they are willing to make, not by what the borrower says about its contracts.

Now scale Barclays' request against that. Insurers covered €4.7 billion of bank credit default risk through synthetic risk transfer in 2025, up from €2.7 billion in 2024, according to a global survey reported by the Financial Times in August 2026. One bank's ask for about $10 billion of cover is roughly twice the entire 2025 synthetic risk-transfer market, which tells you the market for AI credit insurance does not yet exist at the size banks now need.

Read the chain from the top down and the hedge stops being a bank story. Bonds issued to build data centers that CoreWeave will lease already trade tighter than CoreWeave's own bonds, so senior asset-level investors sit in front of the corporate credit, the corporate credit sits behind the banks, and the banks have now sold the default leg to insurers.

That leaves the loss with whoever holds the operating company's paper and, through it, the chips. CoreWeave's trailing depreciation and amortisation of $3.99 billion on $7.59 billion of revenue is the depreciation treadmill in one line: the collateral securing the loans wears out faster than the assets around it. Insurers writing protection on that collateral are underwriting a loss-given-default assumption that depends on second-hand chip prices, an assumption no one has had to test in a real cycle.

The near-term work is mostly disclosure. If Barclays does not name a counterparty or a structure by its February 2027 results, then what is happening is not a completed transfer but a market test the insurance industry has not cleared. A hedge that is never disclosed is a balance-sheet decision, not a market — and how many banks are quietly doing this matters more than the one that talks.

The bear case on AI credit is not that the technology fails. It is that the credit never breaks, in which case insurers earn a premium, banks keep lending and nobody notices. If demand simply outruns the debt, this whole structure is a cheap insurance policy that expires worthless — and the winners are the same people who would have won without it.