NVIDIA and Arm challenge server CPU incumbents

Modal plans to deploy NVIDIA’s Vera CPU for AI agent workloads, expanding beyond infrastructure long based on Intel and AMD processors. The move highlights a broader shift in server computing: CPUs increasingly handle sandbox environments, code execution and tool calls alongside GPU-powered model processing. These workloads place greater importance on memory bandwidth, resource isolation and performance under heavy concurrency.

Vera supports deployment alongside NVIDIA GPUs or in independent CPU servers. Modal’s testing found that it did not consistently outperform competitors on individual tasks, but its throughput advantage increased under concurrent workloads in specific test conditions. Arm is also moving beyond licensing with its AGI CPU, co-developed with Meta, emphasizing rack-level computing density and energy efficiency. Selling processors puts Arm into more direct competition with companies that use its architecture.

AMD is responding through its EPYC processors and Helion infrastructure, which separates agent execution from model inference or combines CPUs and GPUs within a rack. Intel is pursuing high-density computing with Clearwater Forest and higher-performance general computing with Diamond Rapids, while drawing on its established software ecosystem and enterprise relationships. Competition now extends beyond individual processors to control over complete execution platforms. Gartner expects agent-driven CPU demand to grow, but cautions that the expansion will not match the earlier GPU boom and may not primarily benefit traditional CPU suppliers.