European Banks Race to Capture AI Debt Bonanza While Offloading Risk Exposure
N.R. Finch
Europe's biggest banks are rushing to finance AI data centers, then using a tool called SRT to sell the loan-loss risk to outside investors — eating the cake and shedding the calories at the same time.
What are European banks actually doing in the AI boom?
Société Générale disclosed roughly €7.7 billion (about $8.8 billion) in data-center exposure, with around 70% tied to hyperscalers like Amazon Web Services and Microsoft Azure.
BNP Paribas said it is positioning itself to benefit from "the AI capex supercycle," but CFO Lars Machenil was blunt: "We see huge demand, but we have to stay cautious. We're ready to support, but with our eyes wide open."
BBVA and ING are also bundling and offloading AI-linked loan risk; Deutsche Bank said its exposure is "in the high single digits" and stressed it lends only to large, diversified institutions — not small or single-purpose AI firms.
What is SRT — and why is every bank using it?
SRT — significant risk transfer — is essentially insurance on a bank's loan book. The bank seeks protection on 5% to 15% of a portfolio's value, transferring that slice of loss risk to outside investors.
This means → once the risk is off the books, the bank frees up capital to make more loans, pursue M&A, or return cash to shareholders.
In plain terms = the bank earns the lending spread first, then sells the "what if they don't pay us back" piece to someone else — profiting on both ends. The investors who buy that risk? They collect coupon returns above 10% — high yield for high risk.
Who is doing SRT deals, and how big are they?
SocGen is working on an SRT deal worth more than $5 billion, shifting project-finance risk — including data centers — to external investors.
BNP Paribas this month gauged investor appetite for an SRT deal linked to a data-center loan portfolio.
This reflects a broader trend: it is not just European banks. Canada's TD Bank and Royal Bank of Canada are also exploring SRT deals backed by AI infrastructure loans. SRT is moving from niche tool to standard kit for AI financing.
How does Wall Street's approach differ?
U.S. banks prefer to participate in AI buildout through bond underwriting rather than holding large exposures on their own balance sheets — data-center and AI-related loans account for roughly 1% to 1.5% of total lending.
In plain terms = European banks chose "lend, then transfer the risk"; American banks chose "help others issue debt, collect the fee" — one is balance-sheet heavy, the other is a light intermediary model.
Barclays CEO CS Venkatakrishnan said the bank plays a major role in large bond issuances for U.S. hyperscalers and expects to capture more business from adjacent sectors — power, grid infrastructure, and construction.
Where is the real risk in this playbook?
The four largest U.S. hyperscalers have guided combined capex of more than $700 billion this year — that is the total pie European banks are chasing.
This means → the pie is large, but risk is stacking up in parallel: if the AI capex cycle slows or data-center demand disappoints, the risk banks already transferred may not be enough, and what remains on the balance sheet becomes a burden.
Whether SRT keeps working depends on outside investors' willingness to keep buying at reasonable prices — if market risk appetite contracts, this risk-transfer chain could seize up.
Content is for reference only, not financial advice.