← MAIN AI DASH
OCI Zettascale10
Oracle's flagship AI supercomputer connecting hundreds of thousands of NVIDIA GPUs across data centers for extreme-scale AI training.
Visit OCI Zettascale10 →
8.6
Rating / 10
100K daily visits
Overview
OCI Zettascale10 is Oracle's flagship next-generation AI supercomputer, announced in October 2025 as a powerful evolution of the first Zettascale-class cloud computing cluster introduced in September 2024 . The platform delivers 16 zettaFLOPS of peak performance across multigigawatt clusters spanning multiple data centers, integrating up to 800,000 NVIDIA GPUs connected through Oracle's innovative Acceleron RoCE network fabric . The first Zettascale10 supercomputer is deployed in Abilene, Texas, serving as the foundation for the Stargate flagship supercluster in partnership with OpenAI . The platform's power comes from a fundamental rethinking of network architecture. Oracle Acceleron RoCE uses built-in switching on modern GPU NICs, allowing each NIC to connect simultaneously to multiple switches across separate network planes . This creates a flatter, more resilient network fabric that dramatically improves reliability—when problems occur, traffic moves to other planes without expensive checkpoint restarts . The architecture delivers consistent GPU-to-GPU latency across the entire cluster, and reduces network layers, costs, and energy consumption compared to traditional three-tier designs . OCI Zettascale10 was developed in collaboration with OpenAI, with the architecture deployed at the Abilene Stargate site. Oracle is now accepting orders for Zettascale10, with availability targeted for the second half of 2026, supporting up to 800,000 NVIDIA GPUs . The platform represents a new category of AI infrastructure, moving beyond individual supercomputers to interconnected multigigawatt data center campuses purpose-built for the most demanding frontier AI workloads.
✅ Benefits
  • 16 zettaFLOPS peak performance across multigigawatt clusters with up to 800,000 NVIDIA GPUs—unprecedented scale for frontier AI training.
  • Oracle Acceleron RoCE network architecture delivers consistent performance, 2x potential networking capacity, and reliability through isolated network planes.
  • Developed in collaboration with OpenAI and deployed at the flagship Stargate Abilene site, demonstrating production-grade readiness.
⚠️ Drawbacks
  • Targeted at the largest AI workloads—excessive overkill for most enterprise and all individual use cases.
  • Pricing and access are not publicly disclosed; orders require enterprise contracts with significant commitment.
  • Availability is Q2 2026 at the earliest, and initial capacity is likely reserved for marquee customers like OpenAI.