Bank of America has revised upward its medium-term outlook for the server CPU market, projecting growth from about $43 billion in 2026 to $125 billion by 2030, compared with an earlier forecast of $110 billion. The bank’s analysts calculate this implies a compound annual growth rate of roughly 31% over the period, driven by what they describe as a fundamental shift in workload architecture as agentic AI becomes more prevalent.
Analysts led by Vivek Arya say the market is entering a phase where CPUs play a larger role than in the prior GPU-dominated training era. They write that agentic AI - where systems plan, retrieve context, call tools, manage state, and interact with databases - is "structurally even more CPU-intensive," and that CPUs increasingly act as "the control plane of AI inference," responsible for orchestration and memory management across complex multi-step workflows.
Under this view, the expansion of CPU demand represents an enlargement of the overall data center market rather than a one-to-one substitution away from GPUs. BofA expects new rack configurations that are CPU-only to address tasks such as retrieval-augmented generation, vector database operations, agent orchestration, and smaller-model inference. These are workloads the bank says GPU racks previously could not serve as cost-effectively.
As an illustrative example, BofA points to Nvidia’s planned Vera CPU rack, which the bank says is due to launch alongside Nvidia’s Vera Rubin platform in the second half of 2026. The analysts note that a full Vera Rubin pod could approach an approximately one-to-one CPU-to-GPU ratio, reflecting tighter integration between processors and accelerators in future data center architectures.
On market share, BofA projects Intel and AMD will each account for about 28% of server CPU value by 2030. ARM-based custom designs - including offerings such as AWS Graviton, Google Axion, and Microsoft Cobalt - are expected to be the fastest gainers, rising to roughly 37% share from about 15% today. Merchant ARM CPUs are forecast to add another 7% of share. The bank also expects Intel to continue losing share across both cloud and enterprise segments.
Within its vendor preferences, BofA names AMD as its preferred x86 vendor, noting the bank has modestly increased its estimates and retained a $500 price objective for the stock. Nvidia is listed as the other preferred pick, on the basis of its capability to incorporate CPUs into a broader full-stack architecture spanning compute, networking, storage, and memory. BofA assigns Buy ratings to both AMD and Nvidia.
In summarizing its stance, the analysts state that while all major CPU vendors could experience double-digit sales compound annual growth, their top picks are AMD - which the bank expects to maintain cloud share and take share in x86 enterprise - and Nvidia - expected to begin ramping standalone Vera CPUs while integrating CPUs as part of a full-stack approach.
Despite the upgraded trajectory for server CPUs, BofA cautions that these processors will still represent a relatively small portion of total data center systems spending by 2030. The bank estimates server CPUs will account for only about 5-6% of that spending, while AI accelerators remain the dominant component. BofA’s view places AI accelerators at roughly $1.2 trillion of an overall $1.7 trillion AI data center market through 2030.
Implications for markets and sectors
- Semiconductor makers focused on server CPUs stand to benefit from expanded demand driven by agentic AI workloads.
- Data center equipment and cloud services may see shifts in rack architecture and capital allocation toward CPU-optimized configurations for certain inference and retrieval tasks.
- Enterprise IT and cloud providers could adjust procurement and deployment strategies in response to a changing cost-performance trade-off between CPU and GPU racks for specific workloads.
Note: This report reflects the projections and preferences published by Bank of America analysts and does not introduce additional data or timing beyond those cited above.