AMD CEO Lisa Su confirmed on March 3 at the Morgan Stanley Technology, Media & Telecom Conference that demand for server CPUs has surged well beyond what the company had anticipated, driven by the rapid rise of agentic AI applications. The situation is already straining supply chains, and the effects could reach consumers sooner than expected.
For years, GPUs have dominated the AI hardware conversation, with CPUs playing a supporting role. That dynamic is shifting fast. Agentic AI, systems capable of autonomously operating tools, navigating software, and executing multi-step tasks, requires significantly more CPU compute than traditional AI inference workloads. As these deployments have scaled into production environments, demand for server CPUs has exploded at a pace nobody inside the industry fully predicted.
Su was candid about it: she entered this period already holding an optimistic forecast for AMD’s CPU business, projecting annual growth in the high teens percentage range, and even that bullish estimate turned out to be too conservative. AMD’s top cloud customers told her directly that CPU compute demand sitting alongside AI infrastructure had been “under-forecasted” across the board, and that companies are now actively racing to secure supply.
Agent-based AI caught the entire industry off guard
The core issue is that agentic AI workloads put pressure on CPUs in a way that AI training never did. Training is overwhelmingly GPU-bound. Inference, the act of actually running AI models in production and delivering results to users, is more CPU-intensive, and agentic systems compound that demand further by coordinating multiple tools and processes simultaneously.
Su told investors that the CPU-to-GPU ratio in modern AI compute has shifted dramatically in just a few months. The sudden spike in commitments left supply chains without enough time to adjust. AMD says it is actively working with manufacturing partners to ease current bottlenecks, with capacity expansion efforts planned through 2026 and into 2027. Supply tightness is expected to peak around 2026, with gradual improvement heading into 2027.
AMD’s server CPU lineup is positioned well to capture this demand. The company is on track to launch its next-generation EPYC “Venice” processors, based on the Zen 6 architecture and built on TSMC’s 2nm process, in the second half of 2026. Venice has already been confirmed by multiple major cloud providers as a day-one adoption target, and AMD has publicly stated the chips are performing ahead of expectations in the lab.
Zen 6 Ryzen could pay the price
The part that concerns PC enthusiasts is what happens on the consumer side. EPYC Venice and the upcoming desktop Ryzen processors codenamed “Medusa” are both built on Zen 6 architecture using the same TSMC 2nm node. When server demand for that silicon is this high, and margins on data center CPUs are significantly larger, consumer products naturally fall lower on the priority list.
Reports from Taiwanese industry outlet Benchlife, corroborated by multiple hardware publications, indicate that Zen 6 desktop Ryzen processors, originally expected in the second half of 2026, are now being pushed to 2027. The delay would represent a gap of more than 30 months between Zen 5 and Zen 6 for desktop users, the longest break in AMD’s modern CPU generation cadence, which has otherwise held to a roughly two-year rhythm since 2019.
The laptop segment faces similar uncertainty, with the Zen 6-based Medusa Point mobile processors also potentially slipping further. On the competitive front, Intel is reportedly in the same position, its Nova Lake desktop platform is also being pushed to 2027, meaning consumers across both camps are in the same waiting pattern.
There is one piece of good news: Zen 6 desktop CPUs are expected to retain compatibility with existing AM5 motherboards, so when the chips do arrive, upgrading will not require a full platform replacement. But with TSMC’s 2nm capacity stretched thin and data center contracts taking priority, even the 2027 timeline carries uncertainty around availability and pricing at launch.
AMD finds itself in a strong position on the server side, EPYC is gaining ground fast, but that success is creating real friction for the consumer roadmap in the near term.
What do you think, is it fair that server demand keeps pushing consumer CPUs to the back of the line? Tell us in the comments!