With the launch of the Helios initiative at CES, AMD has shifted from selling discrete chips to delivering end-to-end AI solutions. Helios bundles the new MI455 accelerators, EPYC Venice CPUs, high-performance networking fabric and proprietary software into a single rack-scale system. By integrating all components in a pre-validated configuration, AMD now offers customers a plug-and-play AI infrastructure that can be deployed rapidly, reducing integration timelines by up to 30%.

Helios’s modular rack architecture allows hyperscale cloud providers to replicate an identical configuration across thousands of racks. Engineers at a leading North American data center, which has already ordered over 200 Helios racks, estimate that each rack yields approximately 1.2 petaflops of training throughput. This repeatability drives predictable performance per rack and is projected to increase AMD’s AI infrastructure revenue by more than 50% in its first full year of availability.

By tying its top-line growth to system deployments rather than individual CPU or GPU shipments, AMD reduces exposure to the cyclical swings between client processor and data center accelerator demand. In the fourth quarter, system-level orders contributed 35% of server-segment revenue, up from 18% a year earlier. This diversification is expected to narrow quarterly revenue fluctuations from a historical 25% variance to under 10%, providing investors with greater visibility into end-market CapEx trends.