Oracle and OpenAI have cancelled plans to expand their flagship AI data centre campus in Abilene, Texas, after lengthy negotiations broke down over financing arrangements and OpenAI's changing capacity projections, according to reporting from multiple outlets overnight (AEST).
The planned expansion would have grown the facility to as much as 2 gigawatts of capacity, up from the 1.2 gigawatt campus currently under development.The project is developed by Crusoe on Lancium's Clean Campus in Texas, anddevelopment of the 1,000-acre campus remains underway, with multiple facilities already in service, though preliminary agreements to rent a substantial expansion were ultimately dropped.
The breakdown reveals the practical difficulties confronting even the world's largest technology companies as they attempt to build computing infrastructure at an unprecedented scale.Negotiations got complicated due to difficult financing terms and OpenAI's shifting capacity forecasts, which led to their collapse. Lenders and partners have grown cautious about funding multi-billion-dollar AI infrastructure bets when the underlying demand remains uncertain and technology requirements evolve rapidly.
Relations between Oracle and Crusoe have been strained by reliability issues at the site. Earlier this year, data centre buildings went offline for days due to winter weather affecting some of the liquid cooling machinery. These operational problems likely weighed on discussions about committing capital to further expansion.
Critically, the core relationship between the two firms remains intact.Oracle's general partnership with OpenAI remains unchanged. In July last year, Oracle agreed to develop 4.5 GW of data centre capacity for OpenAI, and that program continues. This represents roughly$300 billion over its lifetime.
The failed expansion has nevertheless created an opportunity for a rival.Meta is reportedly in the running to lease the untapped and unbuilt capacity from the site's developer Crusoe after Nvidia stepped in to broker a deal. According to Bloomberg, Nvidia put down a $150 million deposit on the future capacity before approaching Meta about maybe moving in together.
Meta CEO Mark Zuckerberg is happy to burn capital to secure as much data centre capacity as he can lay his eyes on. During the Social Network's Q4 earnings call in January, it announced its intention to plow up to $135 billion into capital expenditures with an eye for GPU compute capacity.
The collapse underscores a broader tension in the artificial intelligence infrastructure race. Tech companies face genuine uncertainty about how much computing capacity they will actually need. OpenAI's shifting forecasts suggest the company itself is not confident in its own demand projections beyond the next year or two. For lenders and infrastructure partners, that uncertainty translates directly into risk;lenders reportedly remain cautious about underwriting debt tied to rapidly evolving technology demand and uncertain long-term pricing for AI compute services.
The practical lesson is sobering. Grand announcements about billion-dollar infrastructure projects sound impressive, but financial discipline requires that companies be honest about what they can actually fund and what demand they can reliably forecast. When negotiators cannot reach agreement on both fronts, projects get shelved, even when both parties would benefit from moving forward. The Abilene setback suggests the AI infrastructure boom, while genuine, may unfold more slowly than the most ambitious early projections suggested.