Oracle unveiled the largest supercomputer for artificial intelligence

With the ongoing boom in artificial intelligence, and about a month after Microsoft announced “the world’s most powerful data center,” Oracle unveiled its OCI Zettascale10, the world’s largest data center.

Oracle has unveiled its OCI Zettascale10, the world’s largest data center.

The supercomputer didn’t come about by accident – it’s the core of the Stargate supercluster that Oracle, along with OpenAI, is building in Abilene, Texas. Zettascale10 is capable of delivering peak performance of up to 16 zettaflops and combining hundreds of thousands of NVIDIA GPUs across multiple data centers into a single, high-performance system.

OCI utilizes Oracle’s Acceleron RoCE network architecture combined with Nvidia’s AI infrastructure to achieve extremely low latency between GPUs. The Zettascale10 clusters will be deployed in high-powered data centers with gigawatts of power consumption, located within a 2 km radius of each other to minimize latency.

OCOI’s Zettascale10 clusters will be deployed in high-powered data centers with gigawatts of power consumption, located within 2 km of each other to minimize latency.

According to OCI executive vice president Mahesh Thiagarajan, this architecture will enable customers to build, train and put large-scale AI models into production, reducing energy costs per unit of computing power and ensuring high system reliability.

Zettascale10 comes about a year after Oracle launched its first Zettascale cloud cluster, unveiled at Cloud World 2024, which could count up to 131,072 Nvidia GPUs. Now the number of GPUs has risen to 800,000, making the new supercomputer one of the largest in the world.

Oracle plans to begin releasing and commercializing Zettascale10 in the second half of 2026. It is the Acceleron RoCE network architecture that is critical to the success of the project, maximizing the efficiency of communication within the clusters.

Oracle’s new Zettascale10 supercomputer will be released and commercialized in the second half of 2026.

Oracle has already invested in OpenAI’s infrastructure, providing 4.5 gigawatts of Stargate capacity worth more than $300 billion. In addition, the company is building new data centers in Texas, New Mexico and the U.S. Midwest.

Oracle is also building new data centers in Texas, New Mexico and the Midwest.

Oracle’s new CEO Clay Magowirk says the rapid expansion of OCI’s infrastructure will help meet the growing demand for efficient and cost-effective training and deployment of artificial intelligence models.