AMD Helios 72-GPU rack with Instinct MI455X and EPYC Venice debuts at Computex 2026 — Supermicro brings hyperscale AMD AI infrastructure to market
Super Micro Computer launched the rack-scale AMD Helios platform at Computex 2026, a 72-GPU double-width system built on AMD Instinct MI455X GPUs, 6th-Gen EPYC Venice CPUs, and Pensando networking — AMD's most aggressive challenge yet to NVIDIA's hyperscale AI rack dominance.
The 72-GPU density is the headline spec. NVIDIA's GB200 NVL72 and the freshly announced Vera-Rubin racks have defined the hyperscale AI rack category for the past year; Helios matches their physical density while running AMD's open ROCm software stack instead of NVIDIA's proprietary CUDA. For cloud service providers, sovereign-AI buyers, and the wave of NeoCloud entrants competing for inference workloads, that opens a viable second source at rack scale — not just at the chip level.
The buyer profile AMD is targeting matters as much as the spec sheet. Helios is positioned for CSPs, hyperscalers, and enterprise customers running LLM training, fine-tuning, and inference. The same week, NVIDIA announced its Vera CPU is in full production with early adoption by OpenAI, Anthropic, and SpaceX — confirming that the frontier-lab tier of compute demand is still NVIDIA-anchored. AMD's bet is the next layer down: the sovereign and NeoCloud buyers who care about open networking, ROCm portability, and avoiding single-vendor lock-in. Whether that segment can absorb 72-GPU racks at the volumes AMD needs is the open question.
Supermicro IR — Supermicro Expands Rack-Scale AI Leadership with AMD Helios Platform → · Yahoo Finance — AMD Helios Rack Platform Puts AI Data Center Ambitions Against Nvidia → · ServeTheHome — AMD EPYC Venice, Instinct MI455X & Helios Hardware On Display →