Morgan Stanley: Agentic AI Shifts Value From GPUs to CPUs and Memory, Creating up to $60bn Incremental CPU TAM by 2030

Why It Matters

The shift redefines where AI spend flows, creating new revenue opportunities for CPU and memory vendors while reshaping semiconductor investment theses.

Key Takeaways

• Agentic AI could add $60 bn CPU market by 2030
• DRAM demand may rise 15‑45 exabytes, 26‑77% of 2027 supply
• CPU‑to‑GPU ratio in AI clusters expected to increase
• Supply‑constrained foundries and interconnects likely to command premiums
• Full‑stack players, not just GPU makers, stand to benefit

Pulse Analysis

The AI landscape is moving beyond the era of single‑task, GPU‑driven models toward autonomous, multi‑step agents that must coordinate reasoning, tool use and persistent context. This orchestration layer relies heavily on general‑purpose processors, turning CPUs into the "control plane" that stitches together model calls, data retrieval and decision logic. As a result, latency in agentic workloads is now dominated by CPU processing, prompting data‑center architects to redesign clusters with higher CPU‑to‑GPU ratios and tighter CPU‑memory integration.

Morgan Stanley quantifies the upside, projecting an incremental $32.5‑60 billion CPU total addressable market by 2030 and a surge of 15‑45 exabytes of DRAM demand—equivalent to up to 77% of the projected 2027 DRAM supply. This translates into a $82.5‑110 billion data‑center market where memory is no longer passive storage but an active participant in continuous learning. The sheer scale of these requirements intensifies pressure on already tight DRAM and advanced packaging supply chains, potentially driving premium pricing for high‑bandwidth memory and low‑latency interconnects.

For investors, the implications are clear: the AI revenue pool is broadening beyond Nvidia‑style accelerators to encompass CPU manufacturers, memory suppliers, advanced substrate producers, and equipment makers. Companies that can deliver end‑to‑end system efficiency—optimizing orchestration, interconnect density, and thermal performance—are poised to capture outsized returns. As the industry rebalances, a diversified exposure across the full stack may outperform a narrow focus on GPUs, reshaping capital allocation strategies in the semiconductor sector.