Morgan Stanley: Agentic AI to Boost CPU and Memory Spending Beyond GPUs
Summary
- • Morgan Stanley projects agentic AI adds $32.5–60B to the data-centre CPU market by 2030
- • Computing bottleneck shifting from GPUs toward CPUs and memory as AI becomes autonomous
- • CPUs are increasingly acting as the control layer for agentic AI multistep coordination
- • GPU demand remains strong — this is expansion of the chip market, not substitution
Details
Agentic AI projected to add $32.5–60B to the CPU market by 2030
Morgan Stanley estimates this incremental demand lands on top of a CPU market already exceeding $100B annually. The wide range reflects uncertainty in agentic AI adoption curves, but even the low end represents a material expansion of addressable CPU spend driven by AI workloads.
Computing bottleneck shifting from GPUs to CPUs and memory as AI turns autonomous
Morgan Stanley's note states: 'As AI transitions from generation to autonomous action, the computing bottleneck is shifting towards CPU and memory, driving a step-change in general-purpose compute intensity.' Agentic systems manage sequential, multistep tasks that require coordination logic better suited to CPUs than to the parallel matrix math GPUs excel at.
CPUs increasingly serve as the control layer for agentic AI orchestration
Unlike inference or training workloads that saturate GPU throughput, agentic AI involves planning, tool use, and iterative decision loops. Morgan Stanley argues CPUs are the natural control plane for this coordination layer, making general-purpose compute a first-class concern in agentic infrastructure design.
GPU demand remains strong — CPU and memory growth is expansion, not substitution
Morgan Stanley explicitly frames the CPU and memory opportunity as additive. The total AI chip market is expanding, not redistributing share away from GPUs. This distinction matters for investors: Nvidia is still named a beneficiary, and the thesis is about a broader spend pool, not a GPU replacement cycle.
Supply-constrained chipmakers could gain pricing power as agentic demand diversifies
As AI spending widens beyond GPUs, companies in bottlenecked parts of the memory and CPU supply chain — particularly those with limited near-term capacity expansion — are positioned to command higher prices. This dynamic could benefit memory suppliers and foundries before new capacity comes online.
Named beneficiaries span CPUs, memory, and chipmaking equipment
CPU and accelerators: Nvidia, AMD, Intel, Arm. Memory: Micron, Samsung, SK Hynix. Manufacturing and equipment: TSMC, ASML. The breadth of this list signals Morgan Stanley views agentic AI as a full-stack infrastructure build-out, not a single-vendor story.
Market Impact = addressable market and competitive dynamics, Insight = analyst thesis, Tech Info = technical architecture point, Strategy = framing of market trajectory, Industry Update = named company developments
What This Means
Morgan Stanley's analysis suggests the AI infrastructure investment thesis is entering a second phase — one where GPUs remain essential but CPUs, memory, and chipmaking capacity become equally critical bottlenecks. For data center operators and AI infrastructure teams, this means planning for a broader silicon footprint as agentic workloads scale. For investors, it reframes the AI chip trade as a full-stack opportunity rather than a GPU concentration bet.