OpenAI and Oracle to Build 64,000 Nvidia GB200 Accelerator Data Center for AI Megaproject Stargate

The transformation of the United States into an AI superpower also involves the creation of a network of data centers across the country as part of the Stargate project, implemented with the participation of OpenAI, SoftBank and Oracle. A data center will be deployed in Texas by the end of 2026 that will house 64,000 Nvidia GB200 accelerators worth billions of US dollars.

According to Bloomberg, a specialized data center with 64,000 Nvidia GB200 accelerators will be built at a site in the Texas city of Abilene by the end of 2026. Stargate’s computing power in Texas will be put into operation in phases, the first of which will already accept 16,000 accelerators of this type. The costs associated with the implementation of this project are not specified, but Nvidia management estimated the older B200 accelerators at between $30,000 and $40,000 each. Thus, the arrangement of just one data center in Texas will cost investors several billion US dollars.

Let us recall that the Stargate project was announced in January by OpenAI, SoftBank and Oracle with the participation of Donald Trump. In the first stage alone, $100 billion is planned to be spent on the development of computing infrastructure in the United States, and over the next four years, the amount may grow to $500 billion. About ten data centers are planned to be built in the United States to service the Stargate infrastructure. In addition to Texas, they may appear in Pennsylvania, Wisconsin, Oregon and Utah.

By the end of 2024, Meta✴ Platforms plans to increase the number of H100 accelerators in operation to 600,000 units. Cloud provider CoreWeave has more than 250,000 Nvidia accelerators distributed across 32 data centers. Elon Musk’s xAI startup has signed a $5 billion contract with Dell to supply servers for its proprietary supercomputer. Such activity by market participants provides large orders for both Nvidia and TSMC, which produces chips for it.