TL;DR — Meta financed a $30 billion Louisiana data center through a joint venture with Blue Owl Capital, keeping the debt off its own balance sheet entirely. Private credit is quietly becoming the default way hyperscalers pay for the AI buildout.
In October 2025, Meta signed a $30 billion financing package for a single data center campus in rural Louisiana. PIMCO is the anchor lender. Blue Owl Capital owns 80% of the entity holding the debt. Meta owns the other 20%, leases the building back once it’s built, and keeps the debt off its own balance sheet entirely. That’s not one company’s clever accounting trick. Across the industry, it’s becoming the default way Big Tech pays for the AI buildout — and it means the debt backing the boom is landing somewhere investors rarely look.
The $3 Trillion Number
Morgan Stanley and Moody’s Ratings put the capital cost of building the data centers the AI industry says it needs at $3 trillion or more over the coming years. JPMorgan’s estimate runs higher — north of $5 trillion once related power infrastructure is included. Last year alone, AI-related companies and projects tapped debt markets for at least $200 billion, and that figure is likely an undercount, since many of the largest deals are private and don’t show up in public bond data. Morgan Stanley expects hyperscalers and their joint ventures to issue another $250 billion to $300 billion in 2026 alone.
None of the biggest technology companies — not Amazon, not Microsoft, not Meta — is willing to fund this out of pocket. The equity stakes flowing into OpenAI and Anthropic don’t come close to covering an infrastructure bill of this size. So the money is coming from debt, and increasingly, from debt structured so it never touches the borrower’s own credit rating.
Why the Debt Disappears From the Balance Sheet
Meta’s Louisiana deal is the clearest template. The financing — $27 billion in debt plus $2.5 billion in equity — sits inside a special purpose vehicle, Beignet Investor, that Blue Owl majority-owns and Meta minority-owns. S&P rates the debt A+. It matures in 2049. Meta acts as developer, operator and eventual tenant of the facility, which will draw up to 5 gigawatts of power and become Meta’s largest data center in the world when it’s finished in 2029. But because Meta doesn’t consolidate the SPV, none of that $27 billion shows up as Meta debt on Meta’s own books.
For lenders, the appeal is a long-dated lease from a company with Meta’s cash flow standing behind the payments. For Meta, the appeal is obvious: build a five-gigawatt campus without moving its own leverage ratios. Expect more of these joint ventures, not fewer — bankers say there’s already a pipeline of roughly $100 billion in similar data center deals waiting to be structured.
Oracle Is Playing a Different, Riskier Version of the Same Game
Oracle’s version skips the joint-venture ownership stake and goes straight to being the tenant. Over the past year, banks have arranged tens of billions of dollars in project finance loans for data centers where Oracle is the intended occupant — including $38 billion in loans for new Wisconsin and Texas facilities being built by Vantage Data Centers as part of Oracle’s Stargate infrastructure contract with OpenAI. The loans are backed by Oracle’s lease commitments, not by a shared ownership structure the way Meta’s are.
That’s a narrower risk concentration. Banks are already grappling with counterparty exposure to Oracle specifically, because the value of those lease-backed loans depends on Oracle continuing to pay rent on facilities built for a single customer relationship — OpenAI’s Stargate buildout. If OpenAI’s own revenue trajectory disappoints, or if Oracle’s Stargate commitments get renegotiated, the loans backing those leases don’t have Blue Owl’s equity cushion or a diversified tenant base to fall back on.
Who’s Actually Holding the Risk
Alphabet, Amazon, Meta and Oracle borrowed a combined $93 billion in the US investment-grade corporate bond market in 2025 — about 6% of all debt issued that year. The $8 trillion investment-grade market absorbed it without much strain. JPMorgan projects roughly $300 billion a year in AI- and data-center-related debt issuance for the next five years, and investment-grade buyers can likely keep absorbing that too, at coupons around 4% to 4.5% for five-year paper.
The riskier money is where it gets more interesting. The US high-yield and leveraged loan market, roughly $3 trillion in total, saw three junk-rated bond deals worth about $7 billion sold last year specifically to finance new data center construction — paying 7% to 9%, nearly double what investment-grade issuers pay. And private credit, the least transparent corner of all, has gone from near-zero AI exposure to more than $200 billion in outstanding loans to AI-related companies in just a few years. Morgan Stanley projects private credit alone will supply an additional $800 billion in data center financing over the next two years.
That’s the part worth sitting with. The hyperscalers funding record capital budgets aren’t writing all the checks themselves anymore. They’re structuring deals so the leverage sits with Blue Owl, PIMCO, private credit funds and whoever buys the next $7 billion junk bond — investors who don’t hold quarterly earnings calls to explain what happens if a data center’s GPUs are obsolete before the debt matures in 2049.
Sources: Bloomberg News reporting by Paula Seligson via EnergyNow (February 2026); Morgan Stanley Research; Moody’s Ratings; JPMorgan estimates; Meta Platforms and Blue Owl Capital joint venture announcement (October 2025, investor.atmeta.com). Third Pole Markets holds no position in Meta, Oracle, Blue Owl or PIMCO as of publication. This is not investment advice — see our About page for our full disclosure policy.
Details on the financing structure are in Meta’s investor relations disclosures and Blue Owl Capital’s own materials on the joint venture.
