Zitron: Stargate Deal Could Kill Oracle's Balance Sheet

The most useful part of Ed Zitron's newsletter piece How OpenAI Kills Oracle is not the headline, it's the ledger he builds underneath it. His central claim is that Stargate, the $500 billion AI infrastructure project announced at the White House on January 21, 2025, is being funded almost entirely by Oracle and its lenders, with no capital coming from OpenAI or SoftBank and no government money behind it.

The numbers he lays out are worth reading slowly. OpenAI has reportedly signed a five-year, $300 billion computing deal with Oracle, sitting alongside a $250 billion Azure commitment, $138 billion with Amazon over eight years, $22.4 billion with CoreWeave, and $20 billion with Cerebras. Zitron adds these up to roughly $852 billion in compute commitments through 2030. In the same window he expects OpenAI to lose more than $167 billion by end of 2028, with the last quarter's free cash flow at around negative $24.7 billion.

On Oracle's side the picture he paints is a company financing the buildout on debt while the tenant is deeply unprofitable. He points to $38 billion of debt taken on for the Wisconsin and Shackelford sites, a $20 billion share sale, and project-financing loans he says keep some of the exposure off the balance sheet. The Abilene site, originally promised for end of 2025, is only partially live in his April 2026 accounting, with two buildings running roughly 96,000 GB200 GPUs against a planned 450,000 at 1.2 gigawatts. He estimates the finished site could cost in excess of $58 billion.

The honest caveat is that Zitron is a professional skeptic writing an opinionated post, not an audited filing. His projected losses, his completion dates, and his read on off-balance-sheet financing are inferences, not disclosures. What the piece does not give you is Oracle's own view of the contract terms, whether OpenAI has minimum-payment guarantees on the $300 billion, or what happens to already-installed hardware if the tenant renegotiates.

If even a fraction of the concentration risk he describes is right, the interesting outcome is not OpenAI's, it's Oracle's, and the ripple would be felt by every hyperscaler that has been asked to underwrite an AI tenant. Track the pace of the Abilene buildout, Oracle's next debt disclosures, and any revision to the five-year contract structure. Those are the numbers that will tell you whether Zitron's ledger holds.