Oracle and OpenAI Cancel Expansion of Flagship Data Center Project

Oracle and OpenAI’s decision to abandon the expansion of the “Stargate” AI data center in Abilene, Texas, reflects a clash between financing expectations and the rapid evolution of OpenAI’s infrastructure needs, Bloomberg reported. Negotiations over the additional capital required for the project stalled, and OpenAI’s shifting workload patterns—driven by the rollout of its latest reasoning models—made the original expansion blueprint less attractive. The fallout opened a window for Meta Platforms, which is now evaluating a lease of the vacant expansion footprint from the site’s developer, Crusoe, according to people familiar with the talks. Nvidia, the dominant AI‑chip supplier, is reportedly acting as an intermediary in Meta’s discussions, a detail confirmed by the same sources who asked to remain anonymous because the negotiations remain private.

The aborted expansion underscores how quickly AI compute demand can outpace long‑term real‑estate planning. When Oracle and OpenAI first announced the Stargate hub in early 2025, the plan called for a multi‑phase build‑out that would eventually house tens of thousands of GPUs to support OpenAI’s growing API traffic. By early 2026, however, OpenAI had introduced its o1‑preview and o1‑mini models—high‑efficiency, iterative reasoning engines that require a different hardware mix than the earlier GPT‑4‑style workloads. Bloomberg notes that these newer models “changed OpenAI’s needs,” prompting the company to reconsider the scale and configuration of the Texas site.

Financial terms also proved a sticking point. Oracle, which was to provide the bulk of the capital and manage the data‑center operations, sought a financing structure that would lock in long‑term revenue streams from OpenAI’s services. OpenAI, meanwhile, grew wary of committing to a fixed‑capacity build as its product roadmap accelerated. The impasse left the expansion site idle, prompting Crusade’s developer, Crusoe, to explore alternative tenants. Meta’s interest, facilitated by Nvidia’s involvement, suggests that the location’s proximity to existing fiber routes and power infrastructure remains valuable, even if the original partnership fell apart.

Industry observers see the episode as a cautionary tale for mega‑scale AI infrastructure projects that hinge on rapidly shifting model architectures. The same Bloomberg piece that reported the cancellation also highlighted that “the collapsed talks created an opening for Meta Platforms Inc. to step in,” indicating that the market for large‑scale AI compute is still fiercely competitive. Nvidia’s role as a broker hints at the chipmaker’s broader strategy to stay embedded in the ecosystem of data‑center deals, leveraging its hardware dominance to shape where and how AI workloads are deployed.

For Oracle, the setback is a reminder that its cloud‑infrastructure ambitions must align tightly with partner expectations. The company has not publicly commented on the cancellation, but Bloomberg’s reporting suggests that Oracle will likely redirect its capital toward other regional data‑center projects where financing terms are clearer. OpenAI, on the other hand, appears to be doubling down on a more flexible, modular compute strategy, favoring partnerships that can adapt quickly to the evolving demands of its next‑generation models. The Abilene episode thus marks a pivot point: massive, monolithic AI hubs are giving way to more agile, partnership‑driven approaches as the industry races toward ever more sophisticated AI capabilities.