Overview
- Oracle and AMD expanded their partnership, with Oracle slated to be the first hyperscaler to offer Helios in a public AI supercluster starting in the second half of 2026.
- AMD indicates initial availability around Q3 2026, aligning Oracle’s launch window with the Instinct MI450 rollout.
- Helios integrates 72 MI450 GPUs per rack with next‑gen EPYC “Venice” CPUs and Pensando “Vulcano” networking, delivering up to 1.4 exaFLOPS FP8, 2.9 exaFLOPS FP4, and 31 TB of HBM4 across the rack.
- Separate reporting describes Oracle’s developing Zettascale10 project that is planned to use up to 800,000 Nvidia GPUs across multiple data centers, including work tied to the Stargate initiative in Abilene, Texas.
- Context for the buildout includes AMD’s recent multi‑year deal to supply OpenAI with AI chips totaling six gigawatts, alongside earlier reports of large Nvidia GPU purchases by Oracle.