AMD unveiled its Helios system at Advancing AI 2026 on July 23, a fully integrated rack-scale solution targeting Nvidia’s dominance in AI infrastructure. Each Helios rack packs 72 Instinct MI455X GPUs alongside 18 6th Gen EPYC Venice server CPUs, delivering up to 2.9 exaflops of peak FP4 performance with 31 TB of HBM4 memory. AMD claims Helios delivers up to 30% more tokens per dollar compared to Nvidia’s Rubin NVL72 rack. Anthropic, OpenAI and Meta have committed to deploying a combined total exceeding 14 GW of MI450-series GPUs in Helios racks. AMD projects the server CPU market will surpass 200 billion dollars by 2030, driven by agentic AI workloads that heavily tax CPUs.
Source: Read the original article
