J.P. Morgan: TSMC’s AI Supply-Demand Gap Hard to Close; Maintain Overweight Rating with NT$2,500 Target Price

Overview

J.P. Morgan released key takeaways from TSMC’s 2026 Annual Shareholders’ Meeting, reiterating its Overweight rating on TSMC with a target price of NT$2,500. The core thesis is that under accelerating AI demand, supply tightness in TSMC’s advanced nodes (N4/N3/N2) will persist through at least 2027 and possibly into early 2028. Capex guidance has room for upward revision, and pricing strategies are turning more proactive, likely supporting resilient gross margins over the coming quarters. Despite competitive pressures and macro uncertainties, TSMC’s structural growth narrative remains robust due to its technological and executional leadership.

Core views

On capex and capacity expansion, management reaffirmed that 2026 capex leans toward $56 billion, with no signs yet triggering a slowdown—indicating investments remain demand-driven. J.P. Morgan expects this guidance to be revised upward further over the next few quarters, with 2027 capex potentially approaching $75–80 billion. The model assumes monthly advanced-node (N5 and below) capacity additions of 90–100k wafers in 2027. Regarding global footprint, while U.S. and Japan fab construction accelerates, Taiwan remains the most efficient R&D and advanced manufacturing hub. By 2028, U.S. capacity is expected to account for only ~10% of total advanced-node capacity, and recent supply shortages have even prompted further local capacity expansion in Taiwan. On AI demand and supply-demand gaps, management noted rising adoption across consumer, enterprise, and sovereign AI segments, driving 2026 revenue growth of over 30%. J.P. Morgan estimates the 2026 N3 supply-demand gap at 600k wafers—the most severe shortage ever recorded. Looking ahead to 2027, as wafer allocation agreements solidify, this gap is unlikely to close soon. Additionally, emerging applications like Agentic AI are generating incremental compute demand, further extending the timeline to supply-demand equilibrium. On advanced packaging and technology roadmaps, regarding market interest in CoPoS technology, management confirmed ongoing engineering development but stated meaningful volume production still requires 2–3 years—consistent with J.P. Morgan’s prior view that scale-up before 2029 or 2030 is unlikely. This implies CoWoS will remain dominant over the next two years, with additional upside potential due to substitution effects. On competition, TSMC emphasized its leadership in advanced foundry and packaging. While Intel remains a top-10 customer, as a competitor it will not receive preferential access to core resources. In contrast, AMD, NVIDIA, and cloud providers’ in-house chips will receive greater allocations of server CPU capacity. On pricing, profitability, and shareholder returns, management indicated wafer pricing should better reflect company value but will adopt a gradual, partnership-based approach to balance long-term client relationships. J.P. Morgan forecasts N3 pricing increases exceeding 10% and N2 increases of 5–10% in 2027. Rush-order pricing for H2 2026 has already risen, while standard N3 pricing remains stable for now. Combined with high utilization rates, these actions are expected to sustain gross margins in the mid-to-high 60% range over the coming quarters. Despite elevated capex, rising free cash flow supports continued dividend growth. Additionally, in response to recent employee dissatisfaction, the company committed to >30% compensation increases in 2026. On mature-node strategy, TSMC clarified it does not plan a phased exit from mature foundry business but will instead focus on high-value specialty nodes—such as CMOS image sensors at Japan’s JASM fab and automotive/industrial chips at Germany’s ESMC fab. J.P. Morgan believes TSMC may exit some low-value-added businesses, but overall 12-inch mature-node capacity will continue growing in the coming years—contrary to market expectations of a broad exit. In this context, Vanguard International Semiconductor (VIS), with its 12-inch capacity expansion and deepening collaboration with TSMC on CoWoS interposers at the VSMC fab, is viewed as a uniquely positioned beneficiary.

Analysis framework

J.P. Morgan’s analysis centers on “AI-driven structural supply-demand imbalance.” First, by dissecting management commentary from the shareholders’ meeting, the report validates the trend of AI demand spreading from training to inference and sovereign AI, and quantifies the N3 supply-demand gap—forming the foundation for assessing capex revisions and pricing power. Second, using generational technology comparison, it distinguishes CoWoS from next-gen CoPoS量产 timelines, refining expectations on the duration of advanced packaging bottlenecks. Third, incorporating geopolitical and efficiency factors, it evaluates the real impact of overseas fab construction on effective capacity, concluding that “Taiwan’s core status remains unchanged.” Finally, translating supply-side tightness into financial forecasts, it uses a dual-driver model (“volume” via utilization + “price” via ASP uplift) to build a closed-loop logic for gross margin improvement and valuation rerating.

Methodology notes

• Industry/ Sector Analysis FrameworkSupply-demand framework

  Quantitative Analysis of Advanced Node Supply-Demand Gaps

  The report goes beyond qualitative descriptions of strong AI demand by specifically estimating a 600k-wafer N3 supply-demand gap. Translating abstract demand into concrete wafer shortage figures is central to determining semiconductor cycle positioning, forecasting capex direction, and assessing pricing power.

• Valuation MethodPE/PEG valuation

  Forward P/E Multiple-Based Pricing

  The NT$2,500 target price is based on approximately 20x 12-month forward P/E. For high-growth tech stocks, forward earnings better reflect future growth expectations than historical earnings; this multiple exceeds the five-year historical average, implying premium recognition of expected gross margin improvement and accelerated growth.

• Competition & Strategy FrameworkMoat / competitive advantage

  Technology Generation Time Window and Switching Costs

  The report emphasizes that competitors need 3–5 years to ramp design and manufacturing to catch up with TSMC. This reflects that the foundry industry’s moat lies not only in current technical specs but also in long customer validation cycles and high switching costs—creating a 'time barrier' that prevents new entrants from gaining share quickly, even if their specs approach parity.

• Company Fundamentals & Financial FrameworkFree cash flow analysis

  Dividend Sustainability Under High Capex

  Despite sharply rising capex, the report remains positive on dividend growth, anchored in the upward trend of free cash flow (FCF). This shows that for capital-intensive firms, dividend capacity should be assessed not by net income or absolute capex alone, but by whether real cash generation—after necessary maintenance and expansion spending—can support shareholder returns.

Asset mapping & comparison

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

• TSMC (2330.TW / TSM.US)

  Core beneficiary: AI demand drives volume and price upside in advanced nodes; capex revisions lock in long-term growth

  Strengths

  Global leadership in technology and execution; extreme N3/N2 supply-demand tightness grants strong pricing power; free cash flow supports dividend growth

  Weaknesses

  High costs of overseas fab construction; geopolitical risks; rising employee compensation pressure

  Comparison

  Compared to Intel, TSMC leads in advanced packaging scale by 10–15x (2028 CoWoS vs. EMIB); compared to mature foundries, focuses on high-value specialty nodes rather than generic capacity

  Risks

  Uncertainty around duration of AI capex cycle; weak PC/smartphone demand in 2H26 could drag non-AI revenue

• Vanguard International Semiconductor (5347.TWO)

  Indirect beneficiary: Deepening collaboration with TSMC on CoWoS interposers; 12-inch capacity expansion fills specific demand gaps

  Strengths

  Rising 12-inch capacity mix; close strategic partnership with TSMC; differentiated positioning in mature-node specialty technologies

  Comparison

  As TSMC focuses on advanced nodes, VIS becomes a unique collaborator—not just a competitor—in the mature-node space

• AMD

  Beneficiary: TSMC allocates more server CPU capacity to AMD, replacing Intel’s share

  Strengths

  Priority capacity support from TSMC; potential server CPU market share gains

  Comparison

  More favorable position than Intel in TSMC capacity allocation

Key data

• 2026 Capex GuidanceApprox. $56 billionManagement reaffirmed leaning toward the high end, with no signs of slowdown
• 2027 Capex Forecast$75–80 billionJ.P. Morgan forecast, significantly revised upward from 2026
• 2026 N3 Supply-Demand Gap600k wafersWidest gap in history, primarily driven by AI demand
• 2027 N3 Price Increase Forecast>10%Reflects value-based pricing strategy, supporting gross margin expansion
• 2027 N2 Price Increase Forecast5%–10%Pricing expectation for next-generation node at launch
• 2026 Employee Compensation Increase>30%Aimed at stabilizing workforce amid recent bonus disputes
• Target P/E RatioApprox. 20xBased on 12-month forward EPS, above five-year historical average

Impact & implications

For TSMC, persistent supply-demand gaps enhance its bargaining power in the supply chain. Higher capex is no longer a burden but a necessary investment to lock in future revenue, and the stock may benefit over the coming quarters from upward earnings revisions and gross margin outperformance. For the industry, the confirmed scarcity of advanced packaging (CoWoS) over the next two years means related supply chains (e.g., substrates, testing) will continue benefiting; meanwhile, mature nodes are not facing broad oversupply, and specialty segments still offer structural opportunities. For competitors, Intel and other challengers cannot meaningfully displace TSMC in the high-end market in the near term, while fabless players like AMD gain a relative advantage through greater capacity allocation.

Risks

• Market divergence exists on the sustainability of the AI capex growth cycle; if downstream monetization underperforms, capex could be scaled back
• If PC and smartphone demand remains weak in 2H26, TSMC’s non-AI revenue could face downside pressure

What to watch

• Whether TSMC’s capex guidance is further revised upward in coming quarters
• Actual N3 shipment volumes and pricing execution in Q3 2026
• CoWoS capacity ramp progress and customer validation feedback
• Strength of PC/smartphone end-market demand recovery in 2H26