Agentic AI workloads are driving a surge in general‑purpose CPU demand, analysts say

Morgan Stanley said agentic AI could add $32.5 billion to $60 billion to data-center CPU demand by 2030 as AI systems shift from answering prompts to carrying out multistep tasks. (finance.yahoo.com) In the bank’s April 20 note, CPUs and memory gained importance as the “computing bottleneck” moved toward coordination, data movement and control, even as demand for graphics processing units stayed strong. (economictimes.indiatimes.com) Morgan Stanley estimated agentic workloads could add 15 to 45 exabytes of dynamic random-access memory demand by 2030, or 26% to 77% of projected 2027 annual supply. (aninews.in) Agentic AI is the industry’s term for software that plans, calls tools and executes sequences of actions, not just a single response. Those systems create more scheduling, memory and input-output work around each model run, which is the kind of work CPUs usually handle. (finance.yahoo.com) (amd.com) AMD said last month that in modern AI clusters, CPUs keep accelerators busy by handling scheduling, data preparation, memory and input-output, and control flow. The company framed agentic inference as a multistep workflow that raises demand for CPU compute. (amd.com) Intel made a similar argument on April 8, when it announced a design with SambaNova that uses Intel Xeon 6 processors as host and “action” CPUs alongside GPUs and SambaNova inference chips. Intel said agentic AI was exposing the limits of GPU-only inference architectures. (newsroom.intel.com) Arm has gone further by branding a new server chip the Arm AGI CPU and describing it as production silicon built for AI infrastructure at scale. Arm said the processor is meant for the expanding CPU role in AI systems, including continuous orchestration and data movement. (arm.com 1) (arm.com 2) Google also underscored the point this month in a joint announcement with Intel, saying CPUs and infrastructure acceleration remain central to AI systems from training orchestration to inference and deployment. (newsroom.intel.com) The immediate effect is not that GPUs are losing relevance. The shift is that AI spending is broadening, with more value moving into the general-purpose chips and memory that coordinate, feed and scale those accelerators. (finance.yahoo.com) (economictimes.indiatimes.com) That is why the new AI build-out is starting to look less like a race for one chip and more like a race for balanced systems. Morgan Stanley’s call, and the chipmakers’ product pitches that followed, are all pointing at the same redesign. (finance.yahoo.com) (amd.com)