OpenAI is no longer planning to expand its Stargate data center partnership with Oracle in Abilene, Texas. The company is building new sites instead, designed around Nvidia's latest hardware generation.
The existing Abilene facility runs on Nvidia's Blackwell GPU architecture and isn't expected to have power running for another year. By that point, OpenAI intends to be deploying Vera Rubin chips, which Nvidia unveiled at CES in January 2026 and is already manufacturing at scale. Vera Rubin delivers roughly five times the inference performance of Blackwell. For a company competing at the frontier of model development, where benchmark positions influence developer adoption and valuations, waiting on a facility built around older hardware isn't a trade worth making.
Oracle pushed back on reports of the breakdown, calling them "false and incorrect," though it stopped short of saying anything specifically about whether expansion was still planned. That careful framing left the door open to the obvious interpretation.
The underlying tension isn't unique to Oracle. Nvidia has compressed its GPU release cycle from roughly two years to one, but data centers still take 12 to 24 months to site, permit, construct, and connect to the grid. Any infrastructure deal signed today is a wager that the hardware it's built around will still be competitive by the time the facility comes online. That wager has become harder to justify.
Oracle's position makes it particularly vulnerable. Unlike Google, Amazon, and Microsoft — each of which funds AI expansion from core businesses generating substantial free cash flow — Oracle has relied heavily on debt. Its balance sheet now carries more than $100 billion in borrowings, and free cash flow has turned negative. Blue Owl, one of Oracle's financing partners on Stargate, is declining to back additional facilities. Oracle enters its fiscal third-quarter earnings with a $50 billion capex plan under investor scrutiny, a stock down 23% year-to-date, and shares more than 50% below their September 2025 peak.
The OpenAI decision is likely an early sign of how quickly infrastructure commitments tied to specific hardware generations can sour. Companies with the balance-sheet flexibility to redirect capital as the hardware landscape shifts are in a very different position from those locked into debt-financed buildouts with multi-year timelines.