Bloomberg’s AI Today newsletter reported on September 23 that the financing cost of the AI buildout — not just its headline size — is now the number investors are watching
The clearest data point: Meta’s $12.5 billion bond offering for its El Paso, Texas data center complex, issued through a special-purpose vehicle called Sopaipilla Investor, priced at yields exceeding 7% in preliminary discussions, with investors demanding roughly a 0.4 percentage point higher risk premium than Meta’s comparable Hyperion data center financing from the previous year.
The structure itself is instructive: the SPV owns 80% of the project, with Meta retaining 20%, and the bonds — maturing in 2048 — are secured by a 20-year lease with Meta effective 2028. It’s a textbook example of how hyperscalers are increasingly financing AI infrastructure off-balance-sheet through asset-backed structures rather than straight corporate debt. A Man Group analysis published earlier this month found total outstanding debt for AI-related borrowers growing at roughly four times last year’s pace, with hyperscaler AI infrastructure spend expected to reach up to $800 billion this year alone — a volume that has made data center securitization one of the fastest-growing corners of structured credit.
Bottom line: The spread widening on Meta’s own paper — one of the most creditworthy names in the sector — is the tell. If investors are asking for more compensation to hold AI data center risk from the biggest, best-capitalized hyperscalers, smaller developers and neoclouds financing similar projects should expect the gap to be wider still, and should be pricing that into project economics now rather than at their next refinancing.
