The planned expansion of Stargate-the massive AI data center in Abilene, Texas-has hit a major roadblock as OpenAI and Oracle have ended their joint project. According to Bloomberg, disagreements over financial terms and shifting power needs from OpenAI led to the collapse of talks. Though construction on the sprawling 1,000-acre campus continues, plans to lease additional capacity have been shelved.
Despite the cancellation of this expansion, the core Stargate infrastructure delivering roughly 1.2 gigawatts (GW) of power is already up and running. Oracle is swiftly deploying Nvidia-based servers that OpenAI uses to train and serve its AI models. However, the cooling systems supplied by Crusoe have experienced disruptions-particularly during winter months-leading to friction between the partners.
Adding to the complexity, Nvidia has stepped in with a $150 million deposit to support Crusoe and is helping scout potential tenants, including Meta*. While Meta is not involved in the Stargate project, it has shown interest in excess capacity at the campus, although no official confirmation has been made by the company yet.
Why the Stargate expansion talks fell apart in Texas
Negotiations to bump up the center’s power from 1.2 to 2.0 GW stalled primarily due to complicated financing and unclear future demand from OpenAI. Local residents have voiced strong opposition to a substantial rise in electricity consumption, likely wary of new gas or coal power plants being brought in to meet the demand. This local pushback created significant headwinds for the project’s growth.
To put it in perspective, 1 GW roughly equals the output of a nuclear power reactor. Yet, unlike some energy-hungry data centers that explore direct connections to nuclear plants, Stargate relies solely on commercial grids and on-site backup generators, which fuels local resistance.
Equipment reliability issues at Crusoe and harsh winter weather also played a role. Cooling system malfunctions rendered some buildings offline for days, shaking confidence between partners. While both OpenAI and Oracle say their collaboration continues, such incidents have cast doubt on the feasibility of a large-scale expansion.
Where Stargate fits in the AI infrastructure race
Stargate stands as one of the largest AI data centers announced in recent years, reflecting the lightning-fast expansion of infrastructure for training massive AI models. Oracle and OpenAI had initially planned a combined capacity of 4.5 GW spread across various sites, including one near Detroit. The cancellation of the Texas extension doesn’t derail the overall trajectory.
Meanwhile, Nvidia is playing an active role, injecting capital and aiding in client acquisition to shore up its ecosystem. But the instability linked to Crusoe’s operations highlights broader industry challenges around scaling complex cooling and power solutions for these mega data centers.
Meta’s* reported interest in capacity beyond Stargate underscores the intensifying competition in the cloud AI services space and the push from major players to diversify their infrastructure sources. How these obstacles will be overcome-and whether new partnership models emerge-remains to be seen in the coming months.
*Meta is recognized as an extremist organization in Russia and banned from operating there.
For the international tech community, this development is significant because it highlights the delicate balancing act between rapid AI infrastructure expansion and the on-the-ground realities of energy supply, local politics, and tech reliability. AI data centers are the backbone of the future of machine learning, and how companies navigate these hurdles will set precedents for scalable AI deployment worldwide. OpenAI’s pivot here, along with Oracle’s strategic positioning and Nvidia’s support, reveal the complex ecosystem of partners driving AI forward-and the growing pains in scaling to match AI’s insatiable compute appetite.