Financing for AI data centres is becoming more expensive as lenders scrutinise loans, sell some exposure and pass risk to insurers. That could make the next wave of computing capacity harder to fund, even as major technology companies continue borrowing heavily.
Watch Desk analysis
What happened
Funding costs for data-centre asset-backed securities are up by more than one percentage point this year, according to figures from Morgan Stanley cited by the CRE Finance Council. Five-year, single-A-rated data-centre securities averaged a spread of 2.57 percentage points over Treasuries in the past three months, compared with 1.95 points in the first quarter.
Issuance of US data-centre asset-backed securities and commercial mortgage-backed securities reached $21.9 billion through 6 November, up 1.2% year on year. But issuance over the latest three months was down 19.9% on the same period last year. Banks are examining deals more closely, and some are selling loans or insuring part of their exposure. The Korea JoongAng Daily report also cites Goldman Sachs’ estimate that hyperscalers issued $296 billion in bonds this year, above its original $250 billion forecast.
Why it matters
Data centres need enormous sums before they can generate revenue. If borrowing costs rise, heavily indebted AI cloud providers may find it harder to finance construction or refinance existing loans. That could slow the infrastructure build-out that AI companies are counting on.
The risk is shifting, not simply vanishing: the report says banks are transferring some loan exposure to insurers. Meanwhile, the largest borrowers are still raising substantial sums. This is evidence of tighter scrutiny and higher costs, not proof that AI investment has stopped or that a financial crisis is imminent.
Our read
The useful signal is the split between a cooler market for some data-centre-backed securities and continued borrowing by the biggest technology companies. The AI build-out has not run out of money, but capital is not equally easy to obtain. Lenders asking harder questions may be healthy discipline; the test is whether sound projects can still get funded without every promise of future demand being treated as collateral.
What to watch
- Whether data-centre borrowing costs and securities issuance continue to move in opposite directions.
- Whether more banks sell loans or insure portions of their AI infrastructure exposure.
- Whether tighter financing changes construction plans or borrowing by AI cloud providers.
Discussion spark: Should lenders keep financing data-centre expansion while AI demand is uncertain, or is tighter scrutiny now the better safeguard against a costly overbuild?
Sources and evidence
- Source update (19 November 2026, 22:55 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.