The $1.3 trillion data center debt club

San Diego-based Atrium, an AI analytics firm, has launched an interactive platform called 'Who Finances America's Data Centers' that tracks the money behind $1.3 trillion and counting of data center investment. The platform's deal-level data shows a capital stack that looks less like a traditional real estate debt market and more like a syndicated loan club, with 38 lenders in one facility, 24 in another, and the five hyperscalers themselves as the biggest borrowers.

Coreweave's $23 billion debt facility is arranged across 38 different lenders, and the $20 billion facility for DigitalBridge and IFM is backed by 24. These are broadly syndicated credit arrangements, not single-lender construction loans, and the borrower side is just as concentrated: Amazon, Microsoft, Google, Meta and Oracle alone account for $223 billion in long-term debt and credit facilities, according to the platform.

Non-bank lenders are a critical part of the mix: PIMCO has done $23 billion in data center originations, a reminder that the private credit machine that has been financing commercial real estate's riskier corners has found a new and seemingly insatiable borrower class. The hyperscalers' appetite for long-term debt suggests they are willing to lock in financing for assets that will take years to build and even longer to cool.

The reported figures don't break out interest rates or maturities, so allocators are left to size the facilities on their own, but the lender counts alone tell a story: spreading a $23 billion facility across 38 institutions diversifies risk while also layering in a coordination problem if construction schedules slip or power costs spike. Construction costs are the other half of that equation: Skanska's Tom Park said the market is seeing 'unprecedented demand in the mechanical, electrical, plumbing and steel space driving up lead times and cost,' a direct threat to any underwriting that assumes a fixed delivery schedule. A data center's economics hinge on power, cooling, and speed; every month of lead-time slippage eats into the yield the 24 or 38 lenders signed up for.

The AI job growth that is redrawing office demand in seven cities is also redrawing the map of data center demand. CBRE's numbers show New York has overtaken San Francisco as North America's biggest tech labor market, with 394,300 tech jobs to 375,730, even though tech is a smaller share of New York's workforce, and Colliers counted AI firms taking 800,000 square feet of New York office space in the second quarter. More AI workers and more office absorption mean more demand for the power-hungry facilities that sit behind the $1.3 trillion number.

None of this makes the $1.3 trillion a bubble; it does make it a logistics problem. The capital is available — the lending clubs prove that — but the constraint is now the supply chain of steel, pipe, and electrical gear, and beyond that the grid itself. As this publication has argued, data centers are power-optionality plays, and the developers who control generation and grid access, not just land, are the ones who will deliver the returns the lenders are underwriting.

The platform's real value for allocators is showing how risk is distributed: a 38-lender club on Coreweave is not the same risk as a 24-lender club on DigitalBridge and IFM, the first a bet on a single borrower and the second a bet on infrastructure sponsors with a portfolio. That distinction is easy to miss in a sector moving at this speed, and it is the kind of transparency the private markets rarely offer. Allocators who want to participate should be reading the facility documents, not the press releases.