AMD and Celestica Collaborate on AI Rack-Scale Platform 'Helios'

AMD and CLS have announced a collaboration to develop Helios, a rack-scale AI platform built on the Open Compute Project's Open-Rack-Wide form factor. The platform will leverage AMD's Ultra Accelerator Link over Ethernet (UALoE) architecture to interconnect next-generation Instinct MI450 Series GPUs, with Celestica handling R&D, design, and manufacturing of scale-up networking switches.

Helios targets cloud, enterprise, and research environments and is expected to become available to customers in late 2026. The open-standard approach positions AMD as a viable alternative to NVIDIA's proprietary NVLink interconnect, appealing to hyperscalers seeking to avoid vendor lock-in. Celestica brings deep expertise in advanced networking switch design, making it a strategic manufacturing partner for AMD's AI infrastructure ambitions.

The partnership signals AMD's push to compete at rack-scale AI deployments, a segment increasingly critical as enterprises move beyond individual GPU servers toward large-scale AI clusters. While NVIDIA's Blackwell architecture remains entrenched, the Helios platform could capture share among cost-conscious enterprise buyers looking for open-ecosystem AI solutions.