Tells Oracle ( ORCL.US) and OpenAI abandon the expansion plan of the Texas AI data center; Meta ( META.US) may take over the lease.

On Friday, sources revealed that Oracle(ORCL.US) and OpenAI have abandoned plans to expand a large artificial intelligence data center in Texas, USA, due to ongoing delays in negotiations over financing arrangements and OpenAI’s constantly changing compute power requirements. This development has opened up opportunities for Meta Platforms(META.US), which is considering leasing the expansion portion of the project.

According to sources, Meta is currently negotiating with project developer Crusoe to lease the expanded data center located in Abilene, Texas. AI chip giant NVIDIA(NVDA.US) has played a key role in facilitating Meta’s contact with the developer to ensure that the future expansion of the data center continues to use its AI chip products rather than those of competitor AMD(AMD.US).

This adjustment highlights the complexity of building ultra-large AI data centers. Such projects typically involve investments of hundreds of billions of dollars and require extensive collaboration among developers, cloud service providers, chip suppliers, and clients.

The Abilene data center campus is part of the highly anticipated “Stargate” project. Announced last year by President Trump at the White House, the project covers approximately 1,000 acres. The campus is still under construction, with some facilities already operational. However, Oracle and OpenAI have ultimately decided not to proceed with the previously discussed expansion plan.

Earlier, the two parties planned to increase the data center capacity from 1.2 gigawatts to about 2.0 gigawatts. Gigawatt-scale data centers are extremely large, with one gigawatt of power capable of generating enough electricity for roughly 750k homes, equivalent to the output of a nuclear reactor.

Sources said the reasons for the breakdown in negotiations include complex financing arrangements and OpenAI’s frequent changes in compute power demand forecasts. Additionally, the data center previously experienced several days of downtime due to winter weather affecting liquid cooling equipment, which also strained relations between Oracle and Crusoe.

Despite this, both sides stated that their partnership remains strong. Oracle said in a statement that the company is proud of its collaboration with Crusoe and the project’s progress, while Crusoe also expressed that both parties are jointly building one of the world’s largest AI compute infrastructure.

Meanwhile, NVIDIA has paid Crusoe a $150 million deposit to ensure its chips continue to be used in the project and has helped push Meta to become a potential tenant for the expansion.

Notably, the collaboration between Oracle and OpenAI has not been significantly affected. Oracle agreed last July to develop 4.5 gigawatts of data center capacity for OpenAI, and related projects are still underway, including other data center projects near Detroit.

As AI model training and deployment demand surge for compute power, the global data center construction cycle has entered an unprecedented expansion phase. Social media giant Meta is heavily investing in AI infrastructure and expects capital expenditures to reach as high as $135 billion by 2026. The company is also building large data centers in Louisiana and Indiana and signed an agreement last month to deploy 6 gigawatts of AMD equipment.

Following these developments, Oracle’s stock price fell 1.18% on Friday to $152.96. Shares of companies involved in AI infrastructure, including CoreWeave, AMD, and NVIDIA, also declined.