Skip to main content

Archived Article — The Daily Perspective is no longer active. This article was published on 9 March 2026 and is preserved as part of the archive. Read the farewell | Browse archive

Technology

Oracle and OpenAI backtrack from Texas expansion, but broader commitment stands

The Abilene Stargate data centre caps at 1.2GW as financing talks collapse, though companies insist their 4.5GW partnership remains intact

Oracle and OpenAI backtrack from Texas expansion, but broader commitment stands
Image: Toms Hardware
Key Points 3 min read
  • Oracle and OpenAI cancelled plans to expand Abilene Stargate data centre from 1.2GW to 2.0GW after financing and demand forecast disputes
  • Both companies insist the broader 4.5GW partnership agreement announced in July 2025 remains on track across multiple US locations
  • Meta is now in talks to lease the cancelled expansion capacity, with Nvidia paying a $150 million deposit to secure the site
  • The cancellation highlights the volatility of hyperscale infrastructure commitments despite record demand for AI compute power

OpenAI and Oracle have halted the expansion of their Abilene, Texas Stargate data centre beyond 1.2 GW due to power grid delays exceeding a year. For months, the two tech giants had been negotiating whether to double the capacity at their flagship facility. Those talks have now collapsed.

The breakdown reveals something investors and policymakers need to understand: even the world's largest infrastructure projects built by some of the most capitalised companies can stumble over practical realities.Oracle and OpenAI 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. This is not a failure of the technology or the vision. It is a failure to reach agreement on who pays for what and when.

Yet here is where the narrative gets muddied, and where companies' public statements matter. On March 9 this year, Oracle issued a statement contesting what it called "incorrect reporting on Stargate expansion."Oracle said it and developer Crusoe are "operating in lockstep" to deliver one of the world's largest AI data centres at the Abilene campus, adding that two buildings are already operational and the remainder of the site is progressing as planned. The company also said it has completed leasing arrangements for an additional 4.5 gigawatts of capacity to support its commitments to OpenAI.

This distinction matters.The Abilene cap represents a constraint on one site within a partnership that extends well beyond a single campus. Oracle committed in July 2025 to a 4.5-gigawatt capacity deal with OpenAI spanning multiple US locations, and current reporting indicates that agreement remains on track. The companies are not backing away from their fundamental infrastructure bet; they are simply not expanding one particular facility.

What derailed the Abilene expansion? Power infrastructure and mismatched expectations.A multi-day outage this year that was caused by winter weather impacting some of the liquid cooling equipment reportedly damaged relations between OpenAI and Crusoe. But more fundamentally,the parties could not agree on capital allocation and financing terms for the expansion, with each side pushing for more favourable economic structures. OpenAI's compute requirements evolved significantly during negotiations as the company iterated on its infrastructure strategy.

The Abilene site itself remains formidable.Bloomberg reported that Oracle leased eight buildings at the Abilene location to house around 400,000 Nvidia Blackwell chips for OpenAI, with only two of those buildings completed so far. Two buildings are operational; the rest are under construction. This is not collapse. This is pause.

Yet the cancellation exposes a truth about hyperscale infrastructure that investors have been reluctant to acknowledge: the bottleneck is not design or ambition, but power and money.With as much as $300bn set to go to Oracle, that cloud provider has itself struggled to fund its data centre buildout, and is loading up on debt and planning to lay off thousands to cover costs. Fellow Stargate backer SoftBank is also turning to debt to fund OpenAI's expansion, and is currently seeking $40bn in loans. This is what financial strain looks like at the largest infrastructure scale.

What happens to the capacity Oracle and Crusoe had planned?GPU giant Nvidia has stepped in to try to help Crusoe lease the remaining capacity to Meta. In an unusual move aimed at ensuring a rival chip designer's product does not end up in data centres at the site, the company has reportedly paid a $150 million deposit to Crusoe to secure the site and is in talks with Meta. This too is revealing. Nvidia's intervention signals that the underlying problem is not demand—which remains ferocious—but the specific commercial terms that made the Abilene expansion unworkable.

The broader Stargate initiative,a $500 billion investment bringing online the flagship site of the $500 billion Stargate program in Abilene, Texas, continues.The companies will continue to work on their 4.5 gigawatt agreement. Additional sites are under development. The infrastructure is still being built.

This story is less about failure and more about the gap between what we announce and what we can execute. Stargate is ambitious—perhaps ambitiously financed. But the fact that one expansion at one campus has been shelved does not mean the partnership or the project has collapsed. It means that real infrastructure, real power, and real money are required to make these ambitions work. That is a lesson worth learning before we commit another half-trillion dollars.

Sources (4)
Grace Okonkwo
Grace Okonkwo

Grace Okonkwo is an AI editorial persona created by The Daily Perspective. Covering the Australian education system with a community-focused perspective, championing evidence-based policy. As an AI persona, articles are generated using artificial intelligence with editorial quality controls.