Agentic AI Reshapes CPU Demand; UBS Raises ARM Target Price to US$245

Overview

This report focuses on the structural impact of Agentic AI on the CPU market. UBS believes that as AI workloads shift from training and traditional inference toward agentic systems featuring orchestration, tool calling, sandboxed execution, file retrieval, and parallel sub-agents, the importance of CPUs is set to increase step-functionally. While the traditional server CPU market continues to grow along historical trends, AI head nodes and standalone CPU servers/racks will create a new layer of demand, driving significant expansion in the CPU market size by 2030.

Core views

Key views include: First, Agentic AI will significantly boost CPU attach rates, benefiting all CPU architectures in the near term; second, as workloads mature, the market will shift from simply securing CPU capacity to prioritizing a combination of throughput, latency, power consumption, memory configuration, and software ecosystem capabilities; third, ARM is likely to be the biggest beneficiary on the server CPU side, while AMD also sees significant upside due to its high core counts, multi-threading, and x86 roadmap, whereas INTC, despite benefiting from the x86 ecosystem and client PC segment, lags relatively in server AI CPUs; fourth, local PC/edge execution may reduce some cloud-side CPU demand but could also catalyze a PC upgrade cycle.

Analysis framework

The report combines expert interviews, historical server CPU market analysis, accelerator shipment and CPU attach ratio assumptions, top-down NVDA AI TAM extrapolation, and an ARM valuation framework. Market sizing employs both bottom-up and top-down methods, utilizing 2030 revenue and pro forma EPS analysis to compare the potential upside elasticity of ARM, AMD, and INTC.

Methodology notes

• market_sizingbottom_up_cpu_tam

  Bottom-Up AI CPU TAM Estimation

  Based on UBS's C2027 US hyperscaler accelerator model, estimating head node CPUs, CPU-per-GPU attach ratios, and standalone CPU rack demand to project 2030 AI CPU unit volumes and revenue scale.

• market_sizingtop_down_ai_tam

  Top-Down AI CPU TAM Estimation

  Starting from NVDA's projected 2030 AI TAM of approximately US$3–4 trillion, back-calculating XPU units and CPU/GPU attach ratios to derive an AI CPU TAM range of approximately US$120B–200B.

• Valuation methodspeg_framework

  PEG Valuation Framework

  Applying a PEG methodology to ARM, raising the target PEG from 1.6x to 1.8x, resulting in a 67x P/E multiple given an approximate 37% long-term EPS CAGR, applied to average 2027/2028E EPS of US$3.68 to arrive at a US$245 target price.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Arm Holdings PLC (ARM.O)

  Primary beneficiary; target price raised and Buy rating maintained

  Strengths

  Power efficiency, low latency, memory efficiency, hyperscaler adoption, NVDA Grace integration, and ARM ecosystem expansion; ~24% 2030E pro forma EPS upside.

  Weaknesses

  Scaling to high core counts is non-trivial, facing challenges related to core interference, software, ecosystem maturity, and licensing/customization constraints.

  Comparison

  UBS believes ARM benefits most on the server CPU side, leading AMD and INTC; projected to reach 40-45% unit share by 2030.

  Risks

  High valuation, pace of software ecosystem maturation, licensing costs, hyperscaler in-house silicon timelines, and uncertainty regarding agentic workload patterns.

• Advanced Micro Devices Inc (AMD.O)

  Key beneficiary of expanding AI CPU demand

  Strengths

  Strong high core counts, multi-threading, x86 software stack, and server CPU roadmap; AMD internally estimates the server CPU market to reach ~US$60B by 2030.

  Weaknesses

  Must compete against ARM's power efficiency and hyperscalers' in-house ARM solutions; some legacy x86 markets remain influenced by INTC.

  Comparison

  UBS ranks AMD's benefit level after ARM but ahead of INTC; ~11% 2030E pro forma EPS upside.

  Risks

  Market share assumptions, changes in AI server architecture, ASP sustainability, competition, and execution risks.

• Intel Corp. (INTC.O)

  Beneficiary but lower in relative ranking; rated Neutral

  Strengths

  Mature x86 ecosystem; per-core performance remains important for tool calling and storage-optimized workloads; potential leverage from AI-driven PC upgrade cycle.

  Weaknesses

  AMD's roadmap leads in the near term, while ARM is stronger in power efficiency and hyperscaler adoption; server AI CPU share under pressure.

  Comparison

  UBS expects INTC to also see tailwinds but ranks it behind ARM and AMD in server CPU benefits; ~7% 2030E pro forma EPS upside.

  Risks

  Coral Rapids execution, market share losses, ASP pressure, process technology, and platform validation risks.

Key data

• ARM Target PriceUS$245Raised from US$175; based on 1.8x PEG, 67x P/E, and average 2027/2028E EPS of US$3.68.
• ARM Current PriceUS$203.26As of 2026-05-04; target price implies ~+21% upside.
• AI CPU TAM~US$120B-200B by C2030ETop-down estimate based on ~40MM XPUs and a 1-to-1 or 2-to-1 CPU/GPU attach rate.
• Bottom-Up AI CPU MarketUS$7B/US$39B/US$125BCorresponding to 2025/2027/2030E.
• Bottom-Up Total CPU TAMUS$31B/US$74B/US$173BCorresponding to 2025/2027/2030E, including both traditional and AI CPUs.
• Traditional Server CPU Growth8% Unit CAGR; 13% Revenue CAGRBased on normalized historical trends from 2005-2020.
• Agentic CPU Attach Increase~3x-8xExpert interviews suggest shifting from traditional training to agentic inference will significantly boost CPU attach rates.
• Agentic CPU Core Demand~80-120 cores/GPUSome experts estimate agentic scenarios could rise from 8-12 cores/GPU for traditional training and 16-24 cores/GPU for inference to 80-120 cores/GPU.
• ARM 2030E Unit Share~40-45%UBS expects ARM's unit share to rise from ~15% exiting 2025 to 40-45% by 2030.
• 2030E Pro Forma EPS UpsideARM +24%; AMD +11%; INTC +7%UBS believes ARM has the highest EPS elasticity.

Impact & implications

If UBS's thesis holds, the CPU investment logic will expand from traditional server cycles to becoming a core component of AI infrastructure. ARM's power efficiency, low latency, and hyperscaler penetration will support server CPU share gains; AMD's high core counts and multi-threading capabilities position it to benefit from standalone AI CPU servers and x86 workloads; INTC can still benefit from the x86 ecosystem, client PC upgrades, and the Coral Rapids catch-up, though it faces relative pressure on the server side. For investors, Agentic AI could lead to a repricing of CPU-related revenues, ASPs, and valuation multiples.

Risks

• Agentic AI deployment is still in early stages; application architectures and cloud/edge division of labor are not yet standardized, potentially resulting in lower-than-expected CPU demand multipliers.
• A shift of more tasks to PCs or edge devices could dampen cloud CPU attach ratio growth; the report notes this could reduce cloud CPU capacity demand by approximately 25%.
• High-core-count CPUs in real-world deployments may face limitations from software scaling, scheduling, memory bandwidth, and power constraints, failing to fully deliver theoretical throughput.
• ARM's high-core-count scaling faces risks related to core interference, SMT maturity, instruction set migration, licensing costs, and ecosystem maturity.
• Macroeconomic factors, international trade, technological substitution, ASP fluctuations, and hyperscaler capex pacing could alter unit volume, revenue, and valuation assumptions.

What to watch

• Whether Agentic AI workloads transition from pilot phases to standardized large-scale deployment.
• Actual CPU/GPU attach ratios, head node configurations, and the pace of standalone CPU rack deployments by hyperscalers.
• Changes in ARM's server CPU share within AWS, GCP, MSFT, and the NVDA ecosystem.
• Next-generation server CPU roadmaps, core counts, frequencies, power profiles, and platform validation results from AMD and INTC.
• Whether local agent execution on PCs/edge devices catalyzes a client CPU upgrade cycle.
• Whether AI CPU ASPs continue to expand alongside increasing core counts and a higher mix of high-end SKUs.