Agentic AI boosts demand for server CPUs, connectivity, and semiconductor equipment

Overview

This report is Citigroup's meeting summary from its Silicon Valley bus tour covering U.S. semiconductor and hardware companies, including management discussions with AMD, AMAT, ALAB, NVMI/Nova, and SMCI. The core conclusion is that Agentic AI is accelerating demand for data center CPUs, connectivity, storage/memory, semiconductor equipment, metrology, and complete systems, with multiple companies across the supply chain reporting stronger demand than previously expected.

Core views

First, Agentic AI is shifting from chat-based inference toward task orchestration, database retrieval, and agent execution, significantly increasing demand for CPUs and server platforms; AMD believes its Agentic share could exceed its share in general-purpose server CPUs. Second, AI data center buildout continues to drive DRAM/HBM, advanced logic, backside power delivery, and advanced packaging, with both AMAT and Nova emphasizing that rising process complexity creates opportunities for equipment and metrology. Third, connectivity and switching are becoming key bottlenecks in scaling AI clusters, and ALAB outlined substantial long-term opportunities around retimers, Scorpio P/X, AEC, CXL, and optical interconnects. Fourth, SMCI emphasized that full-stack data center building blocks and vertical integration help improve profitability and ease supply constraints.

Analysis framework

The report is based on discussions with company management teams and investor relations teams during Citigroup's annual Silicon Valley bus tour. It reviews changes in demand, product roadmaps, supply capacity, competitive landscape, target profitability models, and potential risks by company, and then summarizes the common trends across the AI infrastructure supply chain.

Methodology notes

• Company researchManagement meeting notes

  Validate industry trends through company engagement

  Citigroup held investor meetings with management teams from AMD, AMAT, ALAB, Nova, and SMCI, extracting each company's views on AI demand, product roadmaps, capacity, customers, and competition.

• Market sizeBottom-up opportunity estimation

  Derive TAM from product content and customer demand

  The report cites AMD's $120 billion TAM, ALAB's $25 billion 2030 connectivity opportunity, $10 billion scale-up opportunity, and $4 billion CXL opportunity, reflecting how companies estimate opportunities based on customer demand, content per XPU, and product generation upgrades.

• Profitability modelTarget model and margin framework

  Measure commercialization quality through gross margin, operating margin, and service revenue mix

  ALAB, Nova, and SMCI all discussed target gross margin, operating margin, service revenue, or profit structure, showing that AI demand affects not only revenue growth but also content, product mix, and earnings leverage.

Asset mapping & comparison

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

• AMD

  Management meeting participant; a direct beneficiary of demand for server CPUs and Agentic AI inference.

  Strengths

  Q1 server revenue grew more than 50%; Turin spans 8 to 192 cores; Venice and Verano are positioned as high-performance server products; the company believes CPUs play a critical role in Agentic AI orchestration and execution.

  Weaknesses

  TSMC 3nm lead times remain long; no clear signal yet of AI-driven PC replacement demand; GPU software ecosystem and platform architecture remain competitive focal points.

  Comparison

  AMD believes x86 has advantages over Intel and ARM in security features, multithreading, and broad workload coverage; it also believes CUDA's moat may weaken due to the speed of software iteration.

  Risks

  Bottom-up customer forecasts may be overly optimistic; supply expansion, memory cost pass-through, product mix, and hyperscaler demand timing will affect realization.

• AMAT

  Management meeting participant; a beneficiary of demand for DRAM/HBM, advanced logic process, and advanced packaging equipment.

  Strengths

  Continues to see strong DRAM demand and improving HBM demand; expects systems business to grow more than 30% this year; GAA, backside power delivery, more metal layers, and advanced packaging increase process steps and equipment demand.

  Weaknesses

  The company acknowledges electroplating is not a core strength; execution priorities include supply chain readiness, tool starts, and personnel training.

  Comparison

  AMAT believes DRAM is increasingly following the logic roadmap, and that rising chip complexity expands opportunities for CVD, PVD, CMP, and packaging-related equipment.

  Risks

  The timing of NAND capacity recovery is uncertain; if the supply chain and tool deliveries cannot keep up, revenue realization will be constrained.

• ALAB

  Management meeting participant; a beneficiary of opportunities in AI cluster connectivity, retimers, switching, AEC, CXL, and optical interconnects.

  Strengths

  The company says the 2030 connectivity opportunity is $25 billion; Scorpio P/X corresponds to about a $10 billion switching opportunity; Gen 6 retimers bring a 25%-30% ASP uplift; historical retimer share is about 90%.

  Weaknesses

  Some CXL opportunity has been pushed out due to AI priorities; the automotive retimer opportunity is attractive but not a current focus; part of revenue still depends on customer conversion and design wins.

  Comparison

  In scale-out scenarios, ALAB benefits from Scorpio; NVDA has its own scale-up technology; in AEC, Credo currently has higher share, and switching and retimers also face competition from AVGO, MRVL, and others.

  Risks

  Whether Scorpio P/X customer engagement converts into revenue, the timing of optical interconnects, the duration of copper cable usage, intensifying competition, and customer concentration are the main risks.

• NVMI / Nova

  Management meeting participant; a beneficiary of semiconductor metrology and rising advanced packaging complexity.

  Strengths

  Last year's revenue was $881 million, up 31% year over year, and EPS was $8.62, up 29% year over year; Q1 was a record; the company claims 100% market share in advanced packaging metrology; R&D spending exceeds 15% of revenue.

  Weaknesses

  The company is relatively more conservative than peers, mainly due to visibility considerations; customer adoption still requires matching tool pricing with throughput improvements.

  Comparison

  The more complex GAA, backside power delivery, 3D DRAM, 3D NAND, hybrid bonding, and advanced packaging become, the more favorable it is for metrology companies like Nova.

  Risks

  The WFE cycle, memory price pressure, customer capex timing, and the speed of new architecture transitions will affect growth.

• SMCI

  Management meeting participant; a beneficiary of AI servers, storage, switching, and data center building block solutions.

  Strengths

  Agentic AI is being widely adopted across enterprise, sovereign customers, and neocloud; full-stack data center building block solutions help improve profitability; vertical integration can ease some supply constraints.

  Weaknesses

  In the short term, large customer orders may still have an impact; working capital needs to be optimized to support growth.

  Comparison

  Enterprise customers in industries such as financials and oil and gas care more about system availability than price, and SMCI's customization and integration capabilities help meet that demand.

  Risks

  Customer concentration, supply constraints, working capital pressure, and the delivery timing of large orders may cause revenue and profit volatility.

Key data

• Companies covered in meetingsAMD, AMAT, ALAB, NVMI, STX, WDC, SMCICitigroup said its annual Silicon Valley bus tour covered these companies; this input retains detailed notes for AMD, AMAT, ALAB, Nova, and SMCI.
• AMD server growth>50% in Q1AMD said Q1 server revenue grew more than 50% and that it is seeing a surge in demand driven by Agentic AI.
• AMD server growth drivers70%-75% from unitsAMD expects 70%-75% of server growth to come from shipments, with ASP expansion driven by Turin and a higher-core-count product mix.
• AMD AI opportunity$120bn TAM may be too lowAMD said that if bottom-up customer forecasts are credible, its $120 billion TAM estimate may be too low.
• AMAT systems growth>30% this yearAMAT said its systems business will grow more than 30% this year, with faster growth at advanced NAND/DRAM nodes.
• ALAB long-term opportunity$25bn by 2030ALAB said it solves connectivity problems and sees a $25 billion opportunity by 2030, with the biggest opportunity coming from switching.
• ALAB switching opportunity$10bnALAB said the Scorpio P and X series correspond to an opportunity of about $10 billion, and it expects Scorpio to become its largest product family by year-end.
• ALAB CXL opportunity$4bn by 2030ALAB said the CXL opportunity is about $4 billion by 2030, though AI demand has pushed back part of the deployment timeline.
• ALAB product generation ASP+25%-30%The shift from Gen 5 retimers to Gen 6 typically brings a 25%-30% ASP increase.
• Nova operating metricsrevenue $881M, +31% YoY; EPS $8.62, +29% YoYNova disclosed that both revenue and EPS achieved strong growth last year, and it has more than 7,300 active systems and over 400 customer sites.
• Nova target modelrevenue >$1bn; gross margin 57%-60%; operating margin 28%-33%; EPS $10Nova said it will release a new target model after reaching its current targets.
• SMCI profit structureDCBBS 15%-20% of profitsSMCI expects the profit contribution of its data center building block solutions to rise from the low single digits to 15%-20%.

Impact & implications

The implications of this report for the semiconductor and hardware supply chain are positive: AI demand is spreading beyond GPUs into CPUs, memory, connectivity, advanced packaging, metrology, and complete systems, broadening the range of beneficiaries. However, investment judgment should focus on whether demand can convert into real orders, whether advanced nodes and supply can keep pace, whether competition erodes share, and whether customer concentration and capex timing create volatility.

Risks

• AI demand expectations may be overly optimistic, and customer forecasts may not fully convert into orders and revenue.
• TSMC 3nm, semiconductor equipment lead times, tool starts, and supply chain readiness may constrain near-term supply.
• Competition in connectivity, AEC, retimers, and switching is intensifying, and AVGO, MRVL, Credo, and customer in-house technology may affect market share.
• Long-term opportunities such as CXL and optical interconnects face uncertainty in deployment timing, and AI priorities may delay some general-purpose server architecture upgrades.
• Rising memory prices and cost pass-through may affect customer purchasing cadence, gross margins, and TCO assessments.
• System companies such as SMCI still face risks from concentrated large customer orders, working capital, and delivery execution.

What to watch

• AMD server CPU share, the ramp pace of Turin/Venice/Verano, and hyperscalers' actual purchasing cadence for Agentic AI.
• The sustainability of AMAT and Nova orders related to HBM, DRAM, GAA, backside power delivery, hybrid bonding, and advanced packaging.
• The timing of ALAB Scorpio P/X customer engagement converting into design wins and revenue, as well as penetration of Gen 6 retimers, 800G, AEC, and optical interconnects.
• Whether SMCI's DCBBS profit contribution can rise from the low single digits to 15%-20%, and whether vertical integration truly alleviates supply bottlenecks.
• Technology route choices among CXL, NPO, CPO, and continued copper cabling, and their impact on content opportunities for different suppliers.