At CES, AMD unveiled Helios, a rack-scale AI platform that bundles MI455 GPUs, fourth-generation EPYC Venice CPUs, high-speed networking fabric and software into a repeatable 50-kW pod. Each Helios pod can be replicated across thousands of racks—with OEM partners already contracted to deploy over 5,000 identical systems—enabling hyperscale customers to standardize infrastructure and drive incremental revenue per AI build‐out by up to 20%. By selling complete systems rather than discrete chips, AMD now ties its top‐line growth directly to hyperscaler capital-expenditure plans, smoothing out the traditional volatility between CPU and GPU cycles and creating multi-rack orderbooks for both compute and interconnect components.

Industry analysts highlight the strength of AMD’s EPYC Venice line in data centers as the foundation for the next phase of AI computing. In Q4 last year, the data center segment grew more than 40% year-over-year, driven by EPYC Venice CPU adoption among cloud providers and enterprise HPC clusters. With over 150 design wins announced for AI-optimized servers and forecasts for server CPU unit shipments to exceed 3 million in 2026, AMD’s CPU revenue base now represents a more than 30% share of its total sales mix. This diversification into mainstream PCs and large-scale data centers underpins a longer-term growth runway beyond the cyclical GPU market.

AMD’s latest accelerators emphasize high-bandwidth memory (HBM) over raw compute throughput, targeting inference workloads where memory bandwidth can be the bottleneck. The MI455 GPU, for example, pairs 256 GB of HBM3 across a 6,144-bit interface, delivering over 4.5 TB/s of sustained bandwidth. In early trials, customers reported a 25% performance gain on transformer inference compared to competing devices. With data center GPU revenue projected to grow at a compound annual rate of over 50% through 2026, AMD’s HBM-centric design is expected to win share in inference deployments at major cloud and AI service providers.

AMD’s recently announced 6-gigawatt, multi-year partnership with OpenAI commits both companies to co-develop full-stack AI infrastructure, from custom GPU architectures to turnkey rack deployments. Under the agreement, AMD will deliver 1 GW of Helios-style systems in 2026 and ramp to 3 GW by 2028. OpenAI customers will standardize on AMD-powered clusters, creating an anchor tenant that could generate more than $4 billion in incremental revenue for AMD over the next three years. This collaboration shifts AMD’s narrative from chip supplier to strategic infrastructure partner in the hyperscale AI market, supporting a projected data center revenue trajectory that could exceed $16 billion in fiscal 2024 and approach $34 billion by 2026.