AINEWS
Tech critic Ed Zitron compares the AI buildout to 2008's subprime crisis, pointing to shell company debt stacks that keep most of the leverage off the parent's books, and to one redacted credit agreement whose covenants would decide whether the
CoreWeave disclosed an $8.5 billion debt facility on March 30, 2026, and filed the credit agreement as an exhibit. The dollar figure is plain. The agreement itself is mostly redacted.
The facility sits inside a special purpose vehicle (SPV), a corporate shell created to hold one specific loan. The shell is its own legal entity, with its own assets and creditors, kept separate from the parent company. SPVs are not unusual. Banks have used them for project finance, mortgage-backed securities, and equipment leasing for decades. The point is isolation: a lender can take security over the SPV's specific assets (in CoreWeave's case, GPU clusters and the long-term contracts that rent them) without becoming a general creditor of the parent.
Each SPV carries its own debt, on its own books. Stack many of them under one operator, and the parent company reports only the portion of the debt it has explicitly guaranteed. The rest lives, on paper, somewhere else.
This is the mechanism tech critic Ed Zitron argues is repeating itself in the AI buildout. In a July newsletter titled "The Subprime Datacenter Crisis," Zitron, who writes the "Where's Your Ed At" newsletter and is not a financial reporter, compared the SPV-by-SPV financing structure used by CoreWeave and other AI cloud operators to the synthetic CDOs that amplified losses in 2008. Synthetic CDOs are derivatives that bet on the same pool of mortgage bonds multiple times, so a single default can cascade through the stack. The comparison is opinion, not analysis from a primary lender or regulator, and Zitron has a track record of bearish AI takes. The mechanism he describes, though, is real and visible in the public filings.
A Bloomberg estimate, reported by TechRadar, puts outstanding AI data center debt above $500 billion, with roughly $200 billion held by private credit funds. The aggregate is large enough to be worth a second look at the structure underneath it.
The company closed a $2.6 billion secured debt facility in July 2025 (the so-called DDTL 3.0, its third data center term loan), then added an $8.5 billion facility on March 30, 2026 (DDTL 4.0), disclosed in an 8-K filing with the SEC (the current-report form companies use to announce material events). The credit agreement for one of CoreWeave's borrowing entities, Compute Acquisition Co. VIII, was filed as an exhibit to the same 8-K, with most of the operative terms redacted.
SPV debt can be simple, fully secured, asset-backed project finance with clean covenants, in which case the 2008 parallel doesn't hold. Or it can carry the kind of covenants and structural features that made synthetic CDOs fragile: multiple bets on the same collateral, off-take agreements whose counterparties are the same handful of AI labs, and interest coverage that depends on those contracts holding for a decade. If the redacted credit agreement shows clean covenants, the 2008 parallel breaks. If it shows the features that made synthetic CDOs fragile, the parallel holds.
SPV defaults would land in the same place: the private credit funds that hold roughly $200 billion of the paper. A bad quarter for the AI labs that anchor the off-take contracts would ripple directly into a private-credit portfolio with limited resale options for these loans.
For now, the public signal is in the 10-Q (the SEC's quarterly report form). CoreWeave's next quarterly filing should disclose the aggregate debt held in unconsolidated SPVs and the variable interest entities the parent has to fold back into its own books, the off-take concentration across its top customers, and any covenant ratios tied to those contracts. Read those numbers before the next GPU-order headline, since the leverage sits in the shell-company debt stack that finances the chips.