At CES 2026, AMD introduced the MI455X accelerator built on a 2-nanometer process node and equipped with 432 GB of HBM4 memory per card. By enabling large language models to reside entirely in on-chip memory, the MI455X reduces the need for multi-GPU interconnects and cuts inference system size by up to 40%, translating to lower power draw per rack. Internal benchmarks show the card delivering up to 1.5× higher throughput on common generative AI workloads compared with AMD’s previous-generation MI300X, positioning AMD to capture enterprise customers focused on maximizing compute density and lowering total cost of ownership in data centers.

AMD management reaffirmed a five-year outlook projecting a compound annual growth rate of 60% in its data center segment, driven by accelerating AI infrastructure demand. The company highlighted a tenfold increase in downloads of its ROCm software stack over the past year, signaling growing developer adoption. With NVIDIA indicating cloud GPU supply constraints, hyperscale operators are evaluating AMD’s GPU lineup to fill capacity gaps. Analysts estimate that if AMD captures just 20% of incremental AI data center build-out over 2026–2028, data center revenue could more than triple from current levels.

In January, AMD announced a collaboration with Tata Consultancy Services to co-develop turnkey AI solutions across life sciences, manufacturing and financial services. The agreement covers integration of AMD’s EPYC CPUs, Ryzen AI processors and Instinct GPUs into hybrid cloud and edge architectures. TCS plans to use these platforms to accelerate pilot-to-production transitions for over 50 global enterprise clients, potentially unlocking over $200 million in incremental services revenue by 2027. This alliance underscores AMD’s push to embed its silicon in end-to-end AI stacks and win deals that extend beyond chip supply into managed infrastructure services.