Meta's Newest AI Data Center Cost $27 Billion. None of That Debt Is on Meta's Balance Sheet.
Meta's Hyperion campus in rural Louisiana needed $27 billion to build. Meta didn't borrow it. A joint venture with private-credit firm Blue Owl Capital did — and that one structural trick, legal and now widespread across the AI buildout, is large enough that central banks have started calling it a financial-stability risk.
Whose balance sheet is the $27 billion actually on?
Simplified from the real Hyperion deal: dollar splits for equity differ slightly across reports, so this uses the consistently-reported ~80/20 ownership split and the $27B bond figure. Toggle which structure you're looking at, then press Run to watch the money move.
The plain version
Picture a landlord building you a custom mansion. You don't take out the mortgage — the landlord's investors do. You put up a fifth of the cash to co-own the place, agree to rent it starting the day it's finished, and privately promise to keep paying rent, and cover part of the exit cost, for the next sixteen years even if you don't end up needing it. The bank is thrilled: you're good for it, so the loan is rated almost as safely as if you'd signed it yourself. It just isn't, on paper, your mortgage.
That's roughly what Meta did to build Hyperion, its newest AI data center campus, in rural Richland Parish, Louisiana. The campus needed $27 billion. Instead of Meta borrowing that money and listing it as debt on its own books, Meta formed a joint venture with Blue Owl Capital, a private-credit firm. Blue Owl's investors — fund giant PIMCO put up the most, BlackRock bought over $3 billion of the bonds — now own about 80% of the venture and hold all $27 billion of its debt. Meta owns the other 20%, leases the finished buildings, and has guaranteed it'll keep paying for 16 years. Because of that guarantee, S&P still gave the bonds a solid A+ grade, one notch below Meta's own corporate credit rating, even though none of the $27 billion shows up as Meta's debt.
This trick, perfectly legal and increasingly common, is now how a huge share of the AI data center boom gets built. Private lenders are estimated to be holding well over $100 billion of this kind of debt, up from almost nothing three years ago, and central bankers have started warning it could snap back hard if AI spending ever slows.
The expert version
The mechanism is a joint-venture special-purpose vehicle (SPV) structured to stay unconsolidated under U.S. GAAP's variable-interest-entity rules. Meta's Hyperion campus, a 2-plus-gigawatt AI compute site, sits inside a venture roughly 80%-owned by funds managed by Blue Owl Capital, with Meta holding the remaining 20% equity stake. The venture raised about $27 billion through a privately placed bond offering arranged by Morgan Stanley — PIMCO was the anchor buyer, BlackRock purchased more than $3 billion — with the bonds maturing around 2049.
Meta signs an initial four-year lease on the finished facility, renewable, and separately provides a payment guarantee covering roughly the first 16 years of the venture's debt service, plus a residual-value guarantee tied to the underlying assets. That guarantee is what let S&P rate the bonds A+, one notch below Meta's own corporate issuer rating, despite the debt sitting on the SPV's books rather than Meta's consolidated balance sheet. The accounting rests on Meta not holding a "controlling financial interest" in the entity under ASC 810, even though its guarantee effectively backstops the credit — a structure commentators have compared to project financing long used for pipelines and power plants, now imported into AI compute.
The practice has scaled fast. Industry estimates put private-credit funds' AI data-center debt holdings at roughly $135 billion, up from about $40 billion in 2023, against a separate estimate that AI-related data-center capex could reach $2.9 trillion through 2028, with over half requiring external financing. The Bank for International Settlements' June 2026 Annual Economic Report explicitly flagged this debt-fueled, non-bank-intermediated financing chain as a risk resembling pre-2008 conditions, warning that a hyperscaler capex slowdown could leave AI-supply-chain borrowers unable to service debt. Oracle, which leans on similar leveraged structures for its own AI buildout, saw its five-year credit-default-swap spread hit roughly 2.03 percentage points in July 2026 — an 18-year high — after S&P downgraded its issuer rating to BBB-, one notch above speculative grade.
Why it matters for tech + supply chain: The debt financing AI's physical buildout — chips, power, concrete — increasingly sits off the books of the companies ordering it, which is exactly the structure central banks say could amplify, not absorb, a slowdown in AI spending.
Why it matters for tech + supply chain: Non-recourse-in-name-only SPV debt is becoming the dominant funding rail for AI data centers, concentrating risk in private-credit portfolios that are far less transparent, and far less regulated, than the hyperscalers' own audited balance sheets.