Oracle is making a monumental $40 billion investment in Nvidia's cutting-edge GB200 superchips to fuel OpenAI's next-generation AI infrastructure, according to a Financial Times report. The chips will form the backbone of what could become America's most powerful AI supercomputer at the "StarGate" data center in Abilene, Texas.

Image Source Note: Image generated by AI, authorized service provider MidJourney

The massive purchase covers approximately 400,000 Nvidia GB200 superchips, unveiled at Nvidia's recent GTC event. Each chip combines dual Blackwell GPUs with a Grace CPU, delivering unprecedented processing power. When configured in racks of 36 units (NVL72 systems), they can achieve 1.4 exaFLOPS of sparse FP4 computing power.

At full deployment, Oracle plans to install 11,000 rack systems within a 1.2 gigawatt facility, potentially reaching 16 zettaFLOPs of total computing capacity. However, current power limitations pose significant challenges—the Abilene site currently provides just 200 megawatts, enough for only about 1,500 racks or 54,000 chips.

Power constraints aren't the only hurdle. Each NVL72 rack consumes up to 120 kilowatts, requiring careful energy management across the eight-building campus. Oracle and data center operator Crusoe are exploring solutions, including staggered usage patterns and diversified workloads like inference tasks and synthetic data generation that don't max out system capacity simultaneously.

If successful, the Texas facility would dwarf existing systems like Elon Musk's Colossus supercomputer in Tennessee—potentially offering 10-20 times more computing power. The Colossus currently delivers 800 exaFLOPS using partial power from natural gas turbines while awaiting full electrical infrastructure.

Beyond Texas, OpenAI's StarGate project is expanding globally through partnerships with Oracle, Nvidia, and Cisco. A planned UAE facility is slated to come online in 2026 with an initial 200-megawatt capacity, marking another step in building distributed AI infrastructure worldwide.

Key Points

1. Oracle's $40 billion investment will purchase 400K Nvidia GB200 chips for OpenAI's AI infrastructure
2. The Abilene data center faces power limitations—currently just 200MW available versus 1.2GW planned capacity
3. At full deployment, the facility could deliver 16 zettaFLOPS—surpassing existing supercomputers by an order of magnitude
4. OpenAI plans international expansion with partner facilities including a UAE location launching in 2026