Nvidia pivots to Vera Rubin as Abilene expansion canceled

What to Know:

• Abilene, Texas expansion canceled; Stargate partnership persists alongside Vera Rubin.
• Abilene expansion dropped; Stargate collaboration continues, centered on Vera Rubin.

OpenAI plans to discontinue its cooperation with Oracle’s Texas data center expansion, shifting its core focus to Nvidia’s next-generation chips. The change affects the Abilene Texas data center add-on, while the broader Stargate data center expansion program continues.

According to Tom’s Hardware, Oracle says the main Abilene campus is progressing with two buildings operational and the wider 4.5 GW Stargate capacity agreement with OpenAI remains in force. The decision rebalances near-term buildout toward capacity aligned with Nvidia Vera Rubin rather than extending the Blackwell-based add-on at Abilene.

Why OpenAI avoids mixing Blackwell and Vera Rubin at Abilene

As reported by The Information, OpenAI aims to avoid mixing Nvidia’s Blackwell generation with Vera Rubin at the same site, preferring new or different facilities for Rubin-based deployments. That approach reduces operational complexity and allows designs to be tuned specifically for Rubin-era throughput and efficiency targets.

news/oracle-denies-openai-data-center-delays/?utm_source=openai” target=”_blank” rel=”nofollow noopener”>As reported by Data Center Dynamics in December 2025, Oracle acknowledged supply-chain pressure yet maintained that key deliverables were on track. “Ambitious but achievable,” said Oracle Cloud Infrastructure leadership.

According to TechRadar, several days of winter-weather disruptions to Abilene’s liquid-cooling systems and power delays elevated operational risk for a high-density, mixed-generation footprint. Consolidating new capacity into Rubin-optimized sites is meant to lower those risks and standardize performance envelopes.

As reported by Shacknews, the initial Abilene build targets about 1.2 GW, while an additional ~2 GW expansion is being dropped in favor of Rubin-ready sites. The publication also notes Meta is exploring a lease on the released expansion capacity with Nvidia facilitating discussions.

Hardware roadmap: Nvidia Vera Rubin vs Blackwell, deployment timing

According to AOL’s summary of Nvidia’s guidance, Vera Rubin is moving from prototypes to early samples, with broader deployment expected in late 2026. Analysts highlighted anticipated gains over Blackwell in throughput, energy efficiency, and inference performance, which underpin OpenAI’s preference to start Rubin deployments on clean, purpose-built sites.

In practical terms, aligning build schedules to Rubin’s ramp helps avoid retrofits, cross-generation orchestration overhead, and stranded thermal capacity. The strategy keeps the Stargate program intact while sequencing deployments to the hardware generation that will anchor OpenAI’s next wave of workloads.