Agentic AI drives server CPU TAM up to $170bn+

Overview

Based on the BofA Global Tech Conference and industry discussions, this report reassesses the impact of Agentic AI on server CPU demand. Its core view is that Agentic AI is not merely replacing GPU/XPU workloads, but expanding overall AI system TAM through system complexity, sequential reasoning, tool calling, memory state management, and I/O orchestration, thereby raising CY2030E server CPU TAM to $170bn+.

Core views

The report argues that the role of CPUs in AI infrastructure is shifting from traditional general-purpose compute to system-level orchestration: traditional/IaaS CPUs account for about $30bn, AI head/compute node CPUs about $70bn, and Agentic AI standalone CPU nodes about $70bn. By CY30E, INTC and AMD are each expected to hold about 25% of server CPU value share, ARM merchant about 35%, and ARM custom about 15%. At the stock level, AMD benefits from core count and high-frequency advantages, INTC benefits from long-term CPU and foundry opportunities, ARM benefits from ecosystem share expansion but has a relatively full valuation, NVDA benefits from full-stack system advantages, and QCOM faces intense competition.

Analysis framework

The report uses a combination of TAM reassessment, application scenario breakdown, vendor share forecasting, and valuation re-rating: it first isolates CPU value from the roughly $2.1tn global data center system TAM, then divides demand into traditional/IaaS, AI head/compute nodes, and Agentic AI standalone nodes, and further splits value share and unit share across INTC, AMD, ARM merchant, and ARM custom.

Methodology notes

• TAM estimationThree-way server CPU application framework

  Break down server CPU demand into traditional/IaaS, AI cluster head/compute nodes, and Agentic AI standalone nodes.

  This framework emphasizes that planning, reasoning, retrieval, tool use, and code execution in Agentic AI have low-latency, sequential, and I/O-intensive characteristics, thereby expanding CPU demand rather than simply replacing GPU/XPU.

• Share forecastVendor/architecture value share breakdown

  Forecast CY30 value share across four vendor or architecture categories: INTC, AMD, ARM merchant, and ARM custom.

  The report expects CY30 value share of about 25% each for INTC and AMD, about 35% for ARM merchant, and about 15% for ARM custom, reflecting rapid ARM ecosystem growth in hyperscaler and merchant solutions.

• Valuation methodsPE and SOTP valuation

  AMD uses 42x CY27E PE; INTC uses discounted 25x CY30E PE; ARM uses sum-of-the-parts valuation for its IP and chip businesses.

  The valuation methods reflect differences in profit realization cycles and business structures across companies: AMD focuses on AI CPU/GPU EPS growth, INTC on CY30 IDM profitability, and ARM on the combined value of IP royalty/licensing and its new chiplet business.

Asset mapping & comparison

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

• AMD

  The report views AMD as a core beneficiary on the CPU side and maintains Buy.

  Strengths

  High-frequency CPUs, leading core counts, x86 ecosystem, security and RAS features, potential 256-core EPYC Venice, and opportunities to gain AI CPU/GPU share.

  Weaknesses

  ARM-based processors may accelerate volume ramp from CY27E onward, potentially limiting further AMD share expansion.

  Comparison

  The report says AMD has a modeling advantage versus NVDA Vera and INTC Granite Rapids in high-core-count agentic workloads.

  Risks

  Product launch execution, AI customer conversion, competition with the NVDA ecosystem and new ARM entrants, and valuation sensitivity to high-growth expectations.

• INTC

  The report upgrades INTC twice from Underperform to Buy and raises its target price to $135.

  Strengths

  Expanded server CPU opportunity, supply visibility from in-house manufacturing, foundry opportunities under leading-edge wafer/packaging constraints, and CY30 EPS power raised to $6+.

  Weaknesses

  CY30 value share is expected to decline from about 41% in CY25 to about 24-25%, and execution remains critical.

  Comparison

  INTC is expected to remain the largest by unit share, but must narrow the gap through follow-on products such as Coral Rapids amid high-end performance competition and ARM ecosystem expansion.

  Risks

  Execution risks in advanced process technology, packaging, foundry customer wins, and CPU roadmap delivery.

• ARM

  The report believes ARM is one of the most significant beneficiaries of the server CPU wave, but maintains Neutral.

  Strengths

  Both ARM merchant and ARM custom have share-gain potential, with combined server CPU value share possibly reaching about 50% by CY30.

  Weaknesses

  The stock price above $300 is considered relatively fully valued, limiting near-term upside.

  Comparison

  The ARM ecosystem is being driven by projects such as NVDA Vera, QCOM CPU, AWS Graviton, Google Axion, and Microsoft Cobalt, with faster growth than x86 incumbents.

  Risks

  Commercial chip business monetization, hyperscaler in-house development paths, valuation digestion, and intensifying competition.

• NVDA

  The report maintains NVDA at Buy and calls it the top pick in AI.

  Strengths

  Full-stack AI architecture, tight CPU-GPU-networking integration, Vera Rubin production, strong supply chain visibility, and participation in the agentic CPU opportunity through Vera CPU.

  Weaknesses

  Some CPU frequency and core-count metrics may not necessarily lead AMD's high-end EPYC.

  Comparison

  NVDA's advantage lies not in a single CPU metric, but in system-level synergy, NVLink C2C, ecosystem lock-in, and lowest token cost.

  Risks

  Sustainability of AI capex, system complexity, supply chain expansion, customer in-house ASICs, and competing platform risks.

• QCOM

  The report expects QCOM may unveil a new AI CPU at its Jun-24 AI Day, but maintains Underperform.

  Strengths

  It could re-enter data center CPUs with its Arm-compatible Oryon core and may announce an AI CPU and lead customers.

  Weaknesses

  Existing competitors already have clear advantages, and the report sees its serviceable market as limited.

  Comparison

  Compared with AMD, INTC, NVDA, and existing ARM ecosystem projects, QCOM has a weaker competitive position in data center CPUs.

  Risks

  Uncertainty around customer adoption, product performance, ecosystem compatibility, pricing, and scaling.

Key data

• CY30E server CPU TAM$170bn+Raised from the prior $125bn forecast, implying nearly 5x growth from about $35bn in CY25.
• CY25-30E server CPU TAM CAGR+37%The prior forecast was +29%.
• Global data center system TAM~$2.1tnThe report states that CPUs account for about $170bn of this value.
• AI CPU TAM~$140bnAI head/compute nodes and Agentic AI standalone nodes each account for roughly half.
• Traditional/IaaS CPU TAM~$30bnIncludes traditional CPU workloads in on-premise enterprise and public cloud environments.
• CPU share of data center system TAM8%+Higher than the previous expectation of 5-6%.
• CY30 value share outlookINTC~25%, AMD~25%, ARM merchant~35%, ARM custom~15%ARM's combined value share could approach 50%.
• CY30 unit share outlookINTC~38%, AMD~24%, ARM merchant~19%, ARM custom~18%INTC is still expected to be the largest vendor by unit share.

Impact & implications

The report's implication for the semiconductor sector is broadly positive: if Agentic AI drives more CPU-only racks, head nodes, and high-I/O orchestration demand, server CPUs will evolve from auxiliary components in AI systems into a more important value pool. AMD, INTC, the ARM ecosystem, and NVDA could all benefit, though through different paths; at the same time, GPU/XPU, HBM, networking, and connectivity remain key parts of the overall system, so the report emphasizes expansion of AI system TAM rather than simple CPU replacement of GPUs.

Risks

• Agentic AI demand may materialize below expectations, leading to downward revisions to the $170bn+ server CPU TAM assumption.
• The sustainability of 30%+ CAGR in AI capex is debated; if capex slows, it will affect overall CPU, GPU, and networking demand.
• CPU demand may come more from GPU/XPU substitution rather than system TAM expansion, weakening the report's incremental thesis.
• Volume growth of ARM merchant and custom solutions may compress the market share and ASP of x86 vendors.
• INTC faces execution risk in advanced manufacturing, packaging, foundry customers, and closing the CPU performance gap.
• Valuation re-rating depends on CY27-CY30 earnings assumptions; if EPS or ASP falls short, target price support may weaken.

What to watch

• Progress at AMD Advancing AI 2026 and launches of Venice, Verano, MI455X, and Helios rack.
• Whether QCOM's Jun-24 AI Day announces an AI CPU, lead customers, and a data center roadmap.
• 2H26 mass production, customer adoption, and token cost performance of NVDA Vera Rubin and Vera CPU.
• Progress of INTC Coral Rapids, Diamond Rapids, 14A IP partnerships, Terafab, and external foundry orders.
• Ramp timing of ARM merchant and ARM custom CPUs across AWS, Google, Microsoft, and projects from NVDA/QCOM.
• Whether AI head nodes and Agentic AI standalone CPU racks truly form an approximately 50/50 demand structure.