Oracle and OpenAI have backed out of the planned expansion for their large data center campus in Abilene, Texas, the first major active site of the Stargate project. This decision does not signal the end of either company’s infrastructure plans but reveals something important about the new AI craze: even the most ambitious projects are confronting the realities of power supply, funding, timelines, and rapidly changing demand.
Abilene was set to become a flagship component of OpenAI’s capacity deployment in the United States. Developed by Crusoe on Lancium’s Clean Campus and operated by Oracle for OpenAI workloads, the first phase began in late September 2025. At that time, Crusoe announced that two buildings were operational and that the entire complex would reach around 1.2 GW with eight buildings. The next step was further growth, adding approximately 600 MW to bring the project close to 2 GW. However, this additional expansion will no longer move forward.
Abilene Will Continue Growing, But Not As Originally Expected
What is canceled is not the already committed campus but its additional expansion. According to reports by Bloomberg, which was also cited by Reuters and the Financial Times, Oracle and OpenAI have decided to withdraw that extra capacity in Abilene and shift future capacity to other states. In fact, OpenAI has indicated that they are developing over half a dozen other locations, including a project with Oracle in Wisconsin.
This nuance is important. Abilene is not shutting down: the eight initially planned buildings remain part of the announced deployment, and the campus continues to serve as one of the country’s largest AI complexes. What’s changing is the focus on concentrating even more capacity in this specific location. In other words, Stargate is not disappearing, but its map is starting to be redrawn.
This shift aligns with what OpenAI itself communicated in 2025. In January, it launched Stargate as an initiative aiming to mobilize up to $500 billion over four years to build AI infrastructure in the U.S. Months later, in July, it announced a deal with Oracle to develop an additional 4.5 GW of capacity, and in September, it expanded the plan with five new sites, bringing the total expected capacity to nearly 7 GW. On paper, the message was one of near-unlimited expansion. However, reality is showing that not all megawatts are built at the same pace or in the same places.
AI Race Also Depends on Power Grid and Timing
The cancelation of this expansion highlights how AI infrastructure has become a top-tier industrial issue. Money, demand, and chips are not enough. Land, permits, high-voltage lines, substations, cooling, water, or alternatives to reduce dependence on water, capable contractors, and a resilient supply chain to support rapid construction are also essential.
In the case of Abilene, multiple reports point to this mixture of factors. Bloomberg indicated issues with financing negotiations and frequent changes in OpenAI’s demand forecasts. The Information added that the local electric availability schedule also influenced the decision and that OpenAI would prefer to reserve new capacity for future Nvidia chip generations at other campuses. None of these elements invalidates the value of the Texas project but helps explain why a facility already in operation could suddenly cease to be the preferred site for growth.
This also explains why the industry lives in constant tension between urgency and prudence. Promoters need to reserve energy and capacity years in advance, while clients continuously adjust their plans based on evolving models, costs, and GPU rollout schedules. In this context, a decision like Abilene’s should not be viewed solely as a setback but also as a sign of maturity: large AI contracts are no longer only about headlines but are increasingly focused on identifying locations that can better support future technological waves.
Meta Might Step in to Fill Part of That Capacity
The other consequence of this news impacts power balance among the sector’s giants. With that expansion off the immediate agenda for Oracle and OpenAI, Crusoe needs to find an outlet for the capacity it was aiming to lease. That’s where Meta has emerged as a potential interested party.
According to Bloomberg, Nvidia reportedly facilitated conversations between Crusoe and Meta, aiming for that future capacity to host Meta’s own chips rather than rival hardware. The same reports state that Nvidia had deposited $150 million to secure an option on the site. No agreement has been finalized yet, and several outlets agree that negotiations remain open.
If this move materializes, it would be a highly revealing sign of the current moment: a campus originally designed for Stargate could end up serving the needs of another major AI player, with Nvidia acting not only as a supplier of accelerators but also as an active participant in capacity planning. This wouldn’t be just a real estate deal; it would be another indication that the battle for AI infrastructure is fought both in models and in the data centers that run them.
Ultimately, the Abilene case encapsulates the new landscape well. For months, the prevailing narrative was of near-unlimited expansion of AI infrastructure. But reality is more complex. Projects advance, yes, but not in a straight line. They are relocated, resized, reconfigured, or adjusted to new chip and energy timelines. Stargate is moving forward, but the cancellation of this Texas expansion reminds us that beyond AI enthusiasm, there is a less glamorous but far more critical layer: physical infrastructure, available megawatts, and the true capacity to deliver on promises.
Frequently Asked Questions
Has OpenAI’s Stargate project been canceled?
No. What has been canceled is the planned expansion of the Abilene campus in Texas. The Stargate project continues, and OpenAI is progressing with other developments alongside Oracle in several states.
How much capacity will the Abilene campus have without that expansion?
The Abilene campus remains as previously announced, with eight buildings and an approximate total capacity of 1.2 GW. The additional roughly 600 MW expansion that would have increased capacity near 2 GW is now canceled.
Why is it significant that Meta could enter Abilene?
Because it would demonstrate that data center capacity for AI is so valuable that control can shift rapidly among major tech companies. It also underscores Nvidia’s influence in determining where their GPUs end up.
What does this change reveal about the AI data center market?
That not everything hinges on demand for models. Factors such as financing, available energy, cooling infrastructure, construction timelines, and the cadence of new chip generations are equally crucial. Infrastructure has become the sector’s real bottleneck.