TSMC remains a core beneficiary of AI semiconductor expansion

Overview

This report focuses on TSMC's 2Q26 earnings preview, 3Q26 guidance, advanced process pricing, 2nm/3nm demand, CoWoS and SoIC capacity, as well as the AI GPU, ASIC, CPU, HBM, CPO, and China AI computing supply chain. The core conclusion is that AI semiconductor demand remains strong, cloud CSP capital spending stays elevated, and TSMC holds a core position in the supply chain thanks to advanced process technologies, advanced packaging, and its customer mix.

Core views

The report views TSMC as a key scarce asset in advanced logic foundry: leading technology roadmap and logic density, tight EUV supply, and strong 2nm demand give it the ability to raise prices for leading-edge processes by 5%-10% in 2027. AI demand is driving simultaneous expansion in CoWoS, SoIC, HBM, test equipment, and CPO, but is also squeezing supply and profits in non-AI semiconductors. China's AI computing market is driven by DeepSeek inference demand, improving domestic supply chain capabilities, and local infrastructure advantages, giving domestic AI accelerator TAM high growth potential.

Analysis framework

The report uses top-down TAM estimation, supply chain capacity tracking, customer demand breakdown, cloud capital spending tracking, process roadmap comparison, packaging capacity breakdown, power deployment estimation, and China-US AI chip value chain comparisons to form an integrated view on revenue, capacity, and investment opportunities for TSMC and the AI semiconductor supply chain.

Methodology notes

• Market SizingTAM Scenario Analysis

  Estimate the market size of AI semiconductors, CPU orchestration, and the China AI chip market under bear/base/bull scenarios.

  The report breaks down cloud AI, ASIC, GPU, CPU, HBM, and China AI accelerators separately, and derives market size through 2030 based on assumptions for capital spending, power deployment, chip volumes, and ASPs.

• Supply Chain AnalysisCapacity and Customer Demand Breakdown

  Identify bottlenecks through customer-level breakdowns of CoWoS, SoIC, 2nm/3nm, HBM, and wafer consumption.

  The report highlights demand from customers such as NVIDIA, Broadcom, AMD, AWS, Google, and MediaTek as drivers of TSMC's advanced process and advanced packaging demand.

• Competitive ComparisonTechnology Roadmap and Packaging Solution Comparison

  Compare TSMC, Intel, and Samsung in terms of logic density, process roadmap, and advanced packaging positioning.

  The conclusion favors TSMC, viewing its process roadmap, CoWoS ecosystem, and execution capabilities as still advantageous, though Intel EMIB has potential competition in larger package sizes.

Asset mapping & comparison

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

• TSMC (TSM.US)

  Core beneficiary target

  Strengths

  Leading advanced process technology, strong 2nm demand, scarce CoWoS/SoIC capacity, and the ability to raise prices for leading-edge processes.

  Weaknesses

  Heavy capital spending pressure, while constraints in advanced packaging and EUV supply may limit near-term deliveries.

  Comparison

  Relative to Intel and Samsung, the report believes TSMC still leads in technology roadmap and logic density.

  Risks

  AI demand volatility, slower customer capital spending, advanced packaging competition, geopolitics, and power constraints.

• AI GPU/ASIC supply chain

  Beneficiary chain of demand expansion

  Strengths

  Projects from NVIDIA, Broadcom, AMD, AWS, Google, and others are driving demand for CoWoS, HBM, ABF, and testing.

  Weaknesses

  Some links are constrained by substrates, packaging, HBM, and power.

  Comparison

  In-house ASICs will continue to coexist with NVIDIA GPUs, as CSPs need custom chips to reduce costs and optimize specific workloads.

  Risks

  CSP budget constraints, project delays, supply chain execution risk, and technology roadmap shifts.

• China AI accelerator supply chain

  Opportunity in localization and demand growth

  Strengths

  China AI chip TAM is expected to grow, and domestic chips have lower TCO and better performance-cost advantages in local inference scenarios.

  Weaknesses

  Capabilities in advanced process and high-end packaging remain constrained.

  Comparison

  Compared with NVIDIA processors, domestic chips have better pricing and supply availability in the China market, though there is still a gap in high-end training performance.

  Risks

  Regulatory restrictions, capacity bottlenecks, software ecosystem maturity, and uncertainty in customer order conversion.

• Memory and HBM

  Spillover beneficiary of AI server demand

  Strengths

  AI demand is driving tight supply-demand conditions for HBM, NAND, NOR, and DDR4.

  Weaknesses

  Rising prices may squeeze profits in downstream non-AI semiconductors and equipment.

  Comparison

  HBM has higher strategic importance in AI servers than traditional memory, and NVIDIA still consumes most HBM supply.

  Risks

  Release of new supply, weakening end demand, and an inventory cycle reversal.

• CPO and test equipment

  Beneficiary segment from rising AI/HPC packaging complexity

  Strengths

  Larger package sizes, higher pin counts, and CPO projects are driving demand for test equipment, sockets, and optical engines.

  Weaknesses

  Commercialization pace depends on major customer projects and standard maturity.

  Comparison

  Compared with traditional consumer electronics testing, AI/HPC testing has greater duration and complexity.

  Risks

  Delays in CPO mass production, customer project changes, and intensifying competition.

Key data

• TSMC 2026e Capital SpendingUS$56bnMorgan Stanley's estimate of TSMC capital spending.
• TSMC 2027e Capital SpendingUS$75bnReflects expansion demand in advanced process and advanced packaging.
• Global AI Semi TAM 2030e约US$753bnThe report expects the AI semiconductor market to expand significantly by 2030.
• Global Cloud Capital Spending 2027e近US$1.3tnTop 14 listed global CSPs, excluding sovereign AI.
• Top 4 CSP 1Q26CY Capital Spending YoY+95% Y/YCapital spending by Amazon, Google, Microsoft, and Meta remains strong.
• CPU Orchestration TAM 2030 Base CaseUS$79bnRaised from the previous US$60bn.
• CPU Orchestration TAM 2030 Bull CaseUS$238bnReflects CPU demand elasticity brought by agentic AI.
• China AI Chip TAM 2030eUS$91bnThe report expects China's AI computing demand and domestic accelerator share to increase.
• TSMC 2Q26 EPS Morgan Stanley EstimateNT$25.08Above the consensus NT$24.02 shown in the table.
• TSMC 3Q26 EPS Morgan Stanley EstimateNT$29.21Above the consensus NT$26.98 shown in the table.

Impact & implications

For investors, the report reinforces the structural strength of the AI semiconductor supply chain: TSMC is at the core of advanced process and advanced packaging, while MediaTek, KYEC, ASE, Alchip, GUC, and FOCI benefit from expansion in ASIC, testing, CPO, and packaging; at the same time, memory shortages, tight DDR4/NOR/NAND supply-demand balances, and localization of China AI chips also create niche opportunities. Key constraints include power, budgets, chip capacity, regulation, and the crowding out of non-AI demand.

Risks

• Cloud service provider capital spending or AI budgets may fall short of expectations.
• Power supply, data center construction, and regulatory constraints may limit AI infrastructure expansion.
• Key supply bottlenecks in EUV, CoWoS, SoIC, HBM, and ABF substrates may affect shipments.
• Rising AI semiconductor priority may squeeze demand and profits in non-AI semiconductors.
• Excessively high TSMC capital spending may create depreciation pressure, although the report believes a 60% gross margin should be the bottom.
• Improved execution in Intel EMIB, Samsung advanced processes, or other packaging solutions may bring competitive pressure.
• Localization of China AI chips is affected by advanced process constraints, software ecosystem, sanctions, and uncertainty in customer order conversion.

What to watch

• TSMC 2Q26 results, 3Q26 guidance, and gross margin trend.
• Whether TSMC 2026-2027 capital spending approaches the US$56bn and US$75bn assumptions.
• 2nm customer demand, N3/N5 capacity migration, and the magnitude of leading-edge process price increases.
• Whether CoWoS expands to around 200kwpm by 2027 as expected, and the pace of SoIC expansion.
• Cloud capital spending, capex-to-EBITDA, and AI project deployment among the Top 4 CSPs and Top 14 CSPs.
• Shipments of NVIDIA GB200/300, Vera Rubin, Google TPU, AWS Trainium, and custom ASICs.
• Supply-demand and pricing trends for HBM, DDR4, NAND, and NOR.
• China AI GPU/accelerator orders, TCO performance, TPS performance, and IPO progress.
• Implementation of projects in CPO, PIC optical engines, test equipment, and high pin count sockets.