Oracle plans to purchase around 400,000 Nvidia GB200 chips worth 40 billion dollars. The high-performance processors are to power a new hyperscale data center in Abilene, Texas – with 1.2 gigawatts, one of the largest in the world. Behind the project are OpenAI and SoftBank with their multi-billion-dollar Stargate initiative. The first construction phase is expected to go online in 2026.
The data center is being built as part of the Stargate program, an ambitious infrastructure initiative to strengthen the US AI sector. SoftBank and OpenAI have each pledged $18 billion, while Oracle and the Abu Dhabi state fund MGX have committed an additional $7 billion. Stargate plans to invest up to $500 billion in data centers over the next four years, including internationally - among them a 5-gigawatt site in Abu Dhabi, developed in collaboration with the local provider G42.
The Abilene facility is operated by Crusoe Energy and Blue Owl Capital. They have raised around 15 billion dollars in debt and equity through JPMorgan, including a fresh loan of 7.1 billion dollars. The complex is to comprise eight buildings. Oracle is leasing the site for 15 years, Stargate has not invested directly so far.
With the project, OpenAI is opening a new chapter. Dependence on the previous exclusive partner Microsoft – previously the primary provider of cloud capacities – is being deliberately reduced. The collaboration has been strained recently as OpenAI's demand significantly exceeded Microsoft's offering. An exclusive licensing agreement was lifted at the beginning of the year.
The dimensions set new standards. The Abilene cluster competes with Elon Musk's "Colossus" project in Memphis, which will house a million Nvidia chips in the future. Amazon is also building a data center in Northern Virginia beyond the 1-GW mark with great urgency.
The direction is clear: Whoever wants to remain a leader in training large AI models needs access to massive computing power – and control over its infrastructure.
