AINEWS
Atrium's first of its kind dataset tracks 4,300 facilities and roughly $1.3 trillion in identifiable debt, putting names on the lender side — from asset managers like PIMCO to non bank credit funds and sovereign backed vehicles — for the first time.
A startup called Atrium has put the first names on the lender side of America's $1.3 trillion data-center build. The San Diego AI analytics firm, founded in 2024, published its "Who Finances America's Data Centers" platform on Aug. 18, mapping roughly $1.3 trillion of identifiable debt across 4,300 domestic facilities. The dataset draws on local property records, SEC credit agreements, CMBS, syndicated loan filings, and corporate debt disclosures, a stack of public sources that no one had previously stitched together for the asset class.
The financing side of the AI build had been invisible to outsiders. Capex announcements and analyst notes name the borrowers; the lenders stay behind the agent-bank curtain. Commercial Observer's exclusive on the platform launch is the first public naming pass.
The structure of the credit is what the dataset shows. Coreweave, the GPU-heavy neocloud, has a $23 billion debt facility split across 38 lenders. DigitalBridge and IFM have arranged a $20 billion facility with 24 lenders. PIMCO, the asset manager, has originated roughly $23 billion in data-center loans, a figure that surfaces through Atrium's research rather than a direct PIMCO disclosure. None of these facilities is a single bank holding the risk. The credit has been syndicated, sliced, and resold across the non-bank lending ecosystem.
The non-bank channel is now where most of the new money enters. Titan Investors tallied roughly $120 billion in sovereign-wealth-fund commitments to AI infrastructure in 2026, much of it routed through asset managers and credit funds. DigitalBridge's May 2023 partnership with Saudi Arabia's PIF, which targets Saudi Arabia and the wider GCC, is the prior sovereign-capital datapoint for the asset class.
The borrowers are concentrated. Amazon, Microsoft, Google, Meta, and Oracle together account for roughly $223 billion in long-term debt and credit facilities, per Atrium's aggregation. That figure is a balance-sheet aggregate, not a data-center-only number, and it does not include equity capex. The largest model trainers and platform owners are also among the largest single-name borrowers, and the syndication structures on the Coreweave and DigitalBridge facilities route exposure to that concentration in real time.
The cost side is tightening in parallel. Skanska's Tom Park, a senior construction executive tracking the build, described the project pipeline as a market where "hot and warm may be a better way to say it" in Commercial Observer's Sunday summary. MEP (mechanical, electrical, and plumbing) systems and structural steel are seeing longer lead times. The same demand pull driving the debt cycle is bidding up the physical cost of building.
Two limits on the data are worth flagging. The $1.3 trillion figure is Atrium's proprietary tally of "identifiable" debt, not a settled industry total. Loans that are not in CMBS, not on a public credit agreement, and not disclosed in a syndicated filing do not show up; the actual number is larger and will be revised. The dataset names the lenders but not the concentration of risk among them, including which banks and funds hold the most junior tranches, which hold the senior secured paper, and which are offloading the credit to CLOs and insurance balance sheets.
The next live cases are the 38-lender Coreweave facility and the 24-lender DigitalBridge/IFM deal. Atrium's map, available at atriumdata.ai, will show whether a fifth, sixth, and seventh large non-bank lender joins the league table.