At CES 2026, AMD introduced its next-generation AI hardware portfolio under the “AI Everywhere, for Everyone” banner. Key products include the Helios rack-scale platform and new Instinct MI400 series accelerators. The MI455X and MI440X chips each deliver up to 3 exaflops of AI compute per rack, leveraging CDNA-based architecture and high-bandwidth HBM3 memory. In parallel, the company unveiled Helios, a fully integrated solution combining EPYC Venice CPUs, Instinct MI400 GPUs and Pensando Vulcan NICs to optimize data-center performance for trillion-parameter model training.

AMD’s dual-front approach targets both PC and server markets. On the client side, the Helios rack-scale reference design showcases how Zen-powered Ryzen AI Embedded processors can accelerate on-device inferencing and edge computing workloads. For cloud providers, the MI400 series offers up to 30% higher tensor throughput per watt compared with its predecessor, while maintaining compatibility with the open ROCm software stack. Early partner designs exceed 80% utilization in mixed-precision AI benchmarks, signaling strong interoperability and developer adoption.

Despite Nvidia’s long-standing lead, AMD has narrowed the gap by securing hyperscaler engagements for both MI400 and upcoming MI500 series. The MI500, due in 2027, promises a 1,000× performance uplift over the MI300X through next-gen 2nm process technology and FP4 low-precision formats. AMD’s enterprise value-to-sales multiple stands at roughly half that of its primary rival, reflecting investor confidence in sustained CPU share gains and expanding data-center deployments without the premium typically demanded for closed-ecosystem solutions.