Oracle Commits $40 Billion to AI Infrastructure with Nvidia Superchips

Oracle and Nvidia are teaming up to invest $40 billion in a new AI data center which OpenAI will use to power training and increase global AI infrastructure.

Oracle will invest $40 billion in the infrastructure needed for AI, with its upcoming Stargate data center in Abilene, Texas as its main focus. The recently upgraded facility with 400,000 Nvidia GB200 superchips will allow OpenAI to carry out advanced AI training at major speed.

Every GB200 chip brings together Blackwell GPUs and Grace CPUs and a single NVL72 will make use of 36 GB200 chips for a total performance of 1.4 exaFLOPS. There will be nearly 11,000 racks in Oracle’s new data center and all those racks are expected to reach nearly 16 zettaFLOPS in capacity.

Even so, energy use continues to be a major issue. The plan is for the facility to generate 1.2 gigawatts, but experts say its use could range up to 1.5 gigawatts. The company tries to manage energy-use in part by introducing systems in stages and handling less energy-intensive chores, involving both inference and data processing.

In the beginning, the data center will work at 200 MW in 2025 and move up to its total capacity in 2026. The arrangement which lasts for 15 years indicates that OpenAI will be on the site for a long time.

Such funding also shows how partnerships between OpenAI and Microsoft are shifting. Oracle is strengthening its lead in the AI infrastructure industry as it builds its Stargate AI in the UAE and other countries.