A new report from Morgan Stanley argues the AI value chain is undergoing a structural shift, redirecting capital from a pure GPU race to a full-stack system approach.
The investment narrative for artificial intelligence is moving beyond the singular focus on graphics processing units, according to a new report from Morgan Stanley. The bank argues that as AI evolves from generating content to automating tasks through agents, the industry's primary bottleneck is shifting from raw computation to system-level orchestration, creating a server CPU market worth up to $110 billion by 2030.
"Intelligent agent AI marks a structural shift from computation to orchestration," Shawn Kim, a Morgan Stanley research analyst, wrote in the report. This transition means that while GPUs remain a core component, they will no longer command the entirety of AI budgets and premiums, as a broader set of components becomes critical to performance.
The bank's analysis projects the rise of AI agents will create an incremental market for server CPUs of $32.5 billion to $60 billion by 2030. This is driven by a fundamental change in server architecture, with the CPU-to-GPU ratio moving from a typical 1:12 configuration to as tight as 1:2. The report also forecasts this shift will create 15 to 45 exabytes of new DRAM demand by the same year.
For investors, the report suggests the beneficiaries of AI capital spending will soon expand beyond a handful of chip giants. The next wave of outsized returns may come from "enabling components" in the supply chain that are the first to become bottlenecks and the most difficult to scale, such as advanced substrates, memory, and wafer fabrication capacity.
From Computation to Orchestration
Unlike generative AI, which relies heavily on GPUs for a single, intensive task, AI agents operate through a multi-step workflow. This process involves planning, retrieving data, calling external tools, and iterative refinement—tasks that are inherently better suited for CPUs. Morgan Stanley's core conclusion is that agent-based systems introduce more steps, states, and coordination, elevating the role of the CPU from a supporting component to a mission-critical orchestrator.
This has two major consequences for data center architecture. First, the ratio of CPUs to GPUs will systematically rise. Nvidia's own roadmap suggests a move toward a 1:2 CPU-to-GPU ratio with its Rubin platform, and potentially a 2:1 ratio in future "Rubin Ultra" configurations. Second, DRAM will be upgraded from a simple capacity component to a core driver of system performance and throughput, holding the vast amounts of data needed for context and memory in agent workflows.
New Bottlenecks, New Winners
Morgan Stanley's report identifies several key areas of the supply chain poised to capture value from this architectural shift. The firm is particularly bullish on components with tight supply and long validation cycles, which gives them greater pricing power.
- ABF Substrates: The report sees the current up-cycle for Ajinomoto Build-up Film substrates extending to the end of the decade, with a potential supply-demand gap around 2026-2027. The expanding CPU market alone could add $1.2 billion to the ABF market by 2030.
- Wafer Foundry: Advanced process nodes are critical. The report projects TSMC will increase its share of the CPU foundry market from 70% in 2026 to 75% by 2028 and suggests Intel may begin outsourcing server CPU production to TSMC in late 2027.
- Memory and Interfaces: With DRAM becoming a functional extension of high-bandwidth memory (HBM), companies in the memory hierarchy, including DRAM makers and interface chip providers like Montage Technology, are set to benefit.
While the report sees structural benefits for CPU makers like AMD and Intel, it maintains an "Equal-weight" rating on both, preferring to gain exposure to the agent theme through companies like Nvidia and Broadcom, where it sees a more direct path from capital spending to earnings.
This article is for informational purposes only and does not constitute investment advice.