AMD CEO Highlights Unforeseen CPU Demand Surge from AI Applications
March 6, 2026
The rapid proliferation of agentic artificial intelligence is triggering an unexpected and significant resurgence in demand for central processing units (CPUs) within data centers, a trend that is catching major chip manufacturers off guard and straining supply chains. This shift underscores the evolving compute requirements of modern AI workloads, which increasingly rely on a combination of specialized and general-purpose processors.
During the Morgan Stanley Technology, Media & Telecom Conference this week, AMD Chief Executive Lisa Su revealed that server CPU demand has “far exceeded” the company’s expectations. She attributed this surge primarily to the rise of agentic AI applications, which are capable of autonomous decision-making and task execution. “If you talk to our top customers, they’re like: ‘Wow… Lisa, the demand for CPU compute sitting along AI was perhaps something that was under-forecasted.’ We are in the process of catching up,” Su stated, acknowledging that the situation has tightened CPU supply chains, though AMD is working with customers to alleviate bottlenecks.
The comments follow AMD’s recent launch of its fifth-generation Epyc 8005 Server CPUs, codenamed Sorano. Unveiled ahead of Mobile World Congress in Barcelona, these processors are designed for edge deployments and compute-intensive virtualized radio access network (vRAN) workloads. The new chips support up to 84 Zen 5 cores within a 225-watt power envelope. In a blog post, AMD emphasized the critical role of compute platforms, stating, “As vRAN moves into its next phase of commercial adoption, success will depend on infrastructure choices that balance innovation with operational rigor.”
AMD is not alone in navigating this demand shock. Reports from early February indicated that both AMD and Intel had informed Chinese customers of lengthy wait times for CPUs. Intel’s Chief Financial Officer David Zinsner previously admitted the company misjudged data center CPU demand. While Intel is shifting capacity to address the shortfall, it anticipates the global CPU shortage will persist into its 2026 fiscal year, reaching its lowest level in the first quarter before improving thereafter.
The competitive landscape is intensifying as other tech giants secure future CPU capacity. Last month, Nvidia signed a deal with Meta, which will become the first company to deploy Nvidia’s Grace CPUs at scale, with plans for a large-scale deployment of the forthcoming Vera CPUs in 2027. Shortly after, Meta also entered a separate agreement to become a lead customer for AMD’s upcoming ‘Venice’ and ‘Verano’ Epyc CPUs, the latter slated for 2027. These developments highlight a strategic industry push to secure advanced CPU architectures, which are essential for serial computing tasks that complement the parallel processing power of GPUs used in AI training.
The collective scramble for CPU supply signals a broader industry realization: the infrastructure for advanced AI is more heterogeneous than previously forecasted, requiring robust CPU resources to manage and support agentic systems. This trend is likely to influence data center design, procurement strategies, and chip development roadmaps for years to come.
Source: datacenterdynamics
