Celestica and AMD Announce Collaboration to Advance the Next Era of AI with "Helios" Rack-Scale AI Platform

Executive Summary

• Celestica Inc. and AMD announced a strategic collaboration to develop and bring to market the “Helios” rack‑scale AI platform, combining AMD’s computing leadership with Celestica’s networking switch expertise.
• Celestica will handle R&D, design, and manufacturing of scale‑up networking switches for the Helios architecture, targeting late‑2026 availability.
• The partnership aims to accelerate AI infrastructure deployment across cloud, enterprise, and research environments, improving time‑to‑value and supply‑chain resiliency.

Key Details

• Collaboration Scope: Development of “Helios” rack‑scale AI platform based on Open Compute Project (OCP) Open‑Rack‑Wide (ORW) form factor.
• Celestica’s Role: Responsible for R&D, design, and manufacturing of scale‑up networking switches that will interconnect AMD Instinct™ MI450 Series GPUs via Ultra Accelerator Link over Ethernet (UALoE).
• AMD’s Contribution: Provides high‑performance computing technology, including the MI450 GPU series and overall platform architecture.
• Target Launch: Helios platform expected to be available to customers in late 2026.
• Market Focus: Cloud service providers, enterprise data centers, and research institutions seeking large‑scale AI workloads.
• Strategic Benefits: Faster deployment of AI infrastructure, enhanced performance, improved supply‑chain resiliency, and reduced time‑to‑value for end users.

Notable Quotes

• “Deploying AI at scale requires infrastructure that can be delivered quickly, consistently, and with the performance customers expect,” – Steven Dorwart, Senior Vice President & General Manager, Hyperscalers, Celestica.
• “‘Helios’ represents a new blueprint for AI infrastructure… we are pleased to work with Celestica, leveraging their expertise in delivering leading‑edge networking switch technology,” – Forrest Norrod, Executive Vice President & General Manager, Data Center Solutions Business Group, AMD.