At CES 2026, AMD unveiled its Helios rack-scale AI platform featuring Instinct MI455X accelerators, EPYC Venice CPUs and Pensando Vulcan NICs, along with new Helios and Instinct MI400 series products targeting data-center and PC AI workloads. Despite the technical breadth—Helios promises up to 3 exaflops per rack and support for trillion-parameter model training—the market responded tepidly, sending AMD shares down over 3% in the session following the keynote.

With Helios, MI455X and the MI440X launch, AMD is aggressively expanding its open ROCm ecosystem to challenge Nvidia’s proprietary GPU dominance. The company showcased rack-scale designs capable of matching Nvidia’s NVL72 system in performance, while touting lower inference token costs and greater memory capacity per rack. AMD claims its open-software approach will appeal to hyperscalers seeking alternatives amid rising GPU infrastructure costs.

AMD highlighted deeper relationships with cloud providers at CES, noting multi-year commitments for Instinct hardware from three of the top five global hyperscalers. These commitments underpin AMD’s forecast for AI data-center sales to grow over 60% year-over-year in fiscal 2026. At current multiples, AMD trades at roughly half Nvidia’s enterprise-value-to-sales ratio, offering investors a lower valuation entry into the AI compute build-out.

CEO Lisa Su reiterated AMD’s ‘AI Everywhere, for Everyone’ strategy, betting that embedded AI processors (Ryzen AI Embedded) and new energy-efficient GPU architectures will spark adoption in edge devices, robotics and autonomous systems. The company forecasts AI-PC shipments to exceed 15 million units in 2026, up from 4 million in 2025, and is targeting a doubling of its PC-graphics segment operating margin by year-end.