Morgan Stanley sees agentic AI widening chip spending beyond graphics processors to CPUs

A new report from Morgan Stanley's equity research team highlights a significant shift in the semiconductor landscape driven by the emergence of agentic AI. Unlike traditional AI models focused on single predictions, agentic AI systems are autonomous agents capable of planning and executing complex, multi-step tasks (like coding or data analysis). The analysts argue that this shift in AI architecture will fundamentally change hardware demand, moving it beyond the current dominance of Graphics Processing Units (GPUs) for training and inference.

Morgan Stanley forecasts that agentic AI could drive over $2 billion in incremental spending on Central Processing Units (CPUs) by 2027. The rationale is that while GPUs excel at the parallel matrix math of deep learning, the orchestration, logic, and tool-calling required by AI agents rely more heavily on the serial processing and general-purpose capabilities of modern CPUs. This creates a substantial new growth vector for leading CPU designers like Intel and AMD, even as demand for advanced GPUs from Nvidia and others remains strong.

The report suggests the total addressable market (TAM) for semiconductors in AI is expanding, not just shifting. Agentic workflows will likely run on a combination of hardware, requiring both powerful CPUs to manage agent logic and memory, alongside GPUs or other accelerators for the heavy-lift model inferences. This hybrid compute demand could benefit the broader data center ecosystem, including memory and networking chip makers, as infrastructure evolves to support these more complex AI systems.