A new UBS report forecasts the server CPU market will explode from $30 billion in 2025 to $170 billion by 2030, as the shift to "Agentic AI" workloads reverses the compute-demand hierarchy and places new emphasis on CPU core counts.
"In traditional AI, 70-80% of the compute is on the GPU; in agentic inference, that flips to 70-80% on the CPU," UBS analyst Timothy Arcuri wrote in a May 5 note citing industry expert interviews.
The report finds that while a traditional AI training workload requires just eight to 12 CPU cores per GPU, an agentic AI model needs 80 to 120 cores — a five to 10-fold increase in demand. Following the analysis, UBS raised its price target on ARM to $245 from $175, projecting its market share to climb from 15% to over 40%.
The forecast implies a massive re-rating for CPU-centric companies, with ARM Holdings plc (NASDAQ: ARM) and Advanced Micro Devices, Inc. (NASDAQ: AMD) positioned as the primary beneficiaries. For Intel Corporation (NASDAQ: INTC), the report suggests a tougher battle to maintain relevance in the high-margin AI data center.
Agentic AI Flips the Script on Compute Demand
While the first wave of generative AI was defined by GPU-intensive training, the next phase of "Agentic AI" — where models can autonomously perform complex tasks, use tools, and orchestrate sub-tasks — is fundamentally CPU-bound. According to the report's expert sources, this shift completely changes the data center architecture.
The core reason is that agentic workflows require significant processing for task orchestration, code execution in sandboxed environments, and file retrieval, all of which are CPU-intensive. One expert noted that a single complex AI agent could spawn 10 to 100 sub-agents, each requiring one to four dedicated CPU cores. This dynamic breaks the old "heavy GPU, light CPU" server model and creates a massive new market for high-core-count CPUs.
ARM Set to Capture 50% of AI CPU Revenue
UBS identifies ARM as the clear winner in this architectural shift. The report projects ARM's unit share of the server CPU market will surge from about 15% in 2025 to between 40-45% by 2030. Due to a richer mix of high-priced AI CPUs, its revenue share could reach 50-55%.
The advantage stems from ARM's architecture, which experts claim is about 30% more power-efficient and 20-30% more memory-efficient than competing x86 designs. This has led to its widespread adoption by hyperscalers for their in-house CPU designs, including Nvidia's Grace CPU, AWS's Graviton series, and Google's Axion processors. UBS expects ARM to command over 75% of the "head node" CPU market, which is responsible for orchestrating the GPU racks.
AMD and Intel Face Diverging Paths
For the traditional x86 duopoly, the agentic AI wave presents both opportunity and threat.
Advanced Micro Devices is well-positioned to benefit, according to UBS. The company's strength in high-core-count and multi-threaded CPUs, like its EPYC line, aligns directly with the needs of agentic workloads. The bank's model suggests a potential 11% upside to its 2030 EPS estimates for AMD if the market evolves as predicted.
Intel's situation is more complex. While its x86 architecture will maintain a strong foothold in the traditional enterprise server market, it is rapidly losing ground to ARM in the critical AI head node segment. The report also notes that Intel's upcoming "Coral Rapids" product line is aimed at closing the gap, but for now, AMD and ARM have a more favorable position. UBS sees only a 7% potential upside to its 2030 EPS forecast for Intel, the lowest of the three.
This article is for informational purposes only and does not constitute investment advice.
