AMD and Celestica Partner on Helios Rack-Scale AI Platform for 2026 Launch

Strategic Alliance for High-Performance AI

AMD and Celestica have announced a strategic collaboration to develop and manufacture the Helios rack-scale AI platform on March 16, 2026. This initiative marks a significant step in delivering open-standards infrastructure for large-scale artificial intelligence clusters. Scheduled for release in late 2026, the platform combines AMD leadership in high-performance computing with Celestica specialized networking switch technologies.

Advanced Networking and Open Standards

The Helios architecture is built upon the Open Compute Project (OCP) Open-Rack-Wide form factor. Celestica is leading the research and manufacturing of specialized networking switches designed specifically for this scale-up environment. These switches utilize the Ultra Accelerator Link over Ethernet (UALoE) protocol to ensure high-speed interconnectivity across complex AI workloads.

Powered by Next-Generation Silicon

At the heart of the system lies the upcoming AMD Instinct MI450 Series GPUs. These processors are expected to offer a massive generational leap in memory bandwidth and compute density using 2nm-class fabrication. By integrating 72 of these GPUs into a single rack, the Helios platform can deliver exascale-level performance for training the largest language models.

Industry Adoption and Future Outlook

The collaboration is already gaining traction with major cloud providers across the globe. Oracle Cloud Infrastructure has announced plans to offer a public AI supercluster based on the Helios design starting in late 2026. This partnership aims to reduce time-to-value for enterprises while improving supply chain resiliency across the data center market.

Leadership Perspectives on Innovation

"Helios represents a new blueprint for AI infrastructure," stated Forrest Norrod, executive vice president at amd.com. The companies are working closely to support deployments across cloud, enterprise, and research environments. This move addresses the growing demand for flexible solutions that can handle the most intensive next-generation workloads.