AMD puts Helios rack-scale AI infrastructure into production

AI infrastructureSemiconductorsUnited States

Published: 23 July 2026 (72-hour extension). AMD said Helios, integrating MI455X accelerators, EPYC “Venice” CPUs, networking and ROCm software, is in full production and is expected to ship by the end of the third quarter. reports say that OpenAI expects large deployments from late 2026, with acceleration through 2027.

Why it matters: Competition is shifting from individual GPUs to complete racks, interconnects, cooling, orchestration and serving software. Impact: Buyers gain an alternative to NVIDIA’s vertically integrated stack, but AMD’s “highest performing” description and partner performance claims require independent workload testing. AMD · a report