AMD is experiencing unexpectedly strong demand for server CPUs, with CEO Lisa Su stating that demand has “far exceeded” expectations as agentic artificial intelligence applications accelerate. Speaking at the Morgan Stanley Technology, Media & Telecom Conference, Su described how the CPU-to-GPU ratio in artificial intelligence compute workloads has shifted rapidly in recent months, with major customers reporting that CPU demand adjacent to artificial intelligence accelerators was “under-forecasted.” The sudden spike in customer commitments has tightened supply, and AMD is working with manufacturing and supply chain partners to relieve bottlenecks and expand capacity over the coming year, while positioning its product lineup for training, inference, and agentic workloads.
The changing demand profile is reflected in hyperscaler buying patterns, with companies such as Meta signing CPU agreements with both AMD and NVIDIA, signaling that large cloud providers are diversifying their infrastructure away from an exclusive focus on GPUs. As artificial intelligence systems grow more complex and agentic, CPUs are playing a larger coordinating and data processing role around accelerators. This is driving a more balanced compute architecture, in which high core-count server processors sit alongside cutting-edge GPUs to keep clusters fully utilized and workloads efficiently orchestrated.
The broader industry is feeling similar pressure. Back in late January, Intel disclosed it had to reallocate wafer capacity from its PC division to Xeon production after hyperscalers started ordering larger quantities than anticipated, triggering a temporary shortage. Then, in early February, both Intel and AMD warned customers in China about tightening supplies, with Intel Xeon delivery times stretching to as long as six months and prices up more than 10%. AMD’s situation is somewhat less difficult, but lead times have still pushed out to eight to ten weeks for some orders. On the consumer side, AMD’s next-gen Ryzen “Olympic Ridge” desktop processors based on Zen 6 are rumored to have slipped to 2027, which may indicate that Zen 6 chiplet supply is being prioritized for EPYC server processors as cloud and artificial intelligence demand continues to surge.
