AMD’s CEO Lisa Su has said the company is experiencing server demand that has “far exceeded” expectations.

Speaking during the Morgan Stanley Technology, Media & Telecom Conference earlier this week, Su said that the increased demand had primarily been driven by the rise in agentic AI applications, adding that while this had undoubtedly led to CPU supply chains tightening, AMD was working with customers to ease bottlenecks.

“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,” she said.

Last week, AMD launched its fifth-generation Epyc 8005 Server CPUs, targeting distributed, compute-intensive vRAN workloads, including Layer 1 processing. Unveiled ahead of Mobile World Congress, which took place in Barcelona this week, the CPUs – codenamed Sorano – have been designed for Edge deployments, supporting up to 84 Zen 5 cores while offering a compute density of up to 225W in a power envelope.

“As vRAN moves into its next phase of commercial adoption, success will depend on infrastructure choices that balance innovation with operational rigor. Compute platforms play a central role in that balance,“ AMD said in a blog post outlining the CPUs.

The CPU resurgence

AMD is not the only company grappling with an unexpected increase in demand for CPUs, with Su’s comments following reports from early February that both AMD and Intel had also informed Chinese customers wanting to purchase CPUs that they would face lengthy wait times.

Intel CFO David Zinsner also previously admitted that the company misjudged demand for data center CPUs, and while it was now "shifting as much [capacity] as we can over to the data center,” the chipmaker still expects the global CPU shortage to continue into FY2026, reaching its “lowest level” in Q1 before improving in Q2 and beyond.

Nvidia, meanwhile, signed a deal with Meta last month that will see the social media giant become the first company to deploy Nvidia’s Grace CPUs at scale. Under the terms of the agreement, Meta is also targeting an additional “large-scale deployment” of Nvidia’s forthcoming Vera CPUs in 2027. Nvidia announced it would be making Vera available as a standalone offering in January of this year, with neocloud CoreWeave to be the first company to get its hands on the chips.

The following week, Meta signed a separate deal with AMD that will see it also become a “lead customer” for the chip designer’s ‘Venice’ CPUs and its ‘Verano’ Epyc CPUs, the latter of which are targeted for 2027.

CPUs are processors with a limited number of cores, making them best suited for serial computing tasks. GPUs, by comparison, have thousands of smaller cores and are used for parallel computing tasks, such as AI and high-performance computing.