TSMC Full Investment Analysis
Report date: 2026-04-27
Type: Initial position
Core conclusion: Buy / core holding; 12-month target NT$2,100 (ADR ~$220)
Executive summary
TSMC (NYSE: TSM / TWSE: 2330) is in a historic window overlapping an "AI super cycle + process-technology monopoly + pricing-power surge." In Q1 2026 gross margin reached 66.2% and net margin topped 50.5%, with advanced process (≤7nm) contributing 74% of wafer revenue and the HPC platform rising to 61% as the largest platform.
The core of the framework is the "pricing-power / capacity paradox": 2nm wafer ASP exceeds $30K, capacity is only one-third of customer demand, and CoWoS advanced packaging is sold out through 2027. Samsung's 3nm yield disaster (20-40%) and Intel 18A's external revenue of just ~$300M leave TSMC in an effective monopoly at the 2nm node.
The certainty has a time boundary — 2028 is the key inflection year: by then Samsung 2nm may mature, Intel 14A may reach production, CoWoS may be in surplus, A14 (1.4nm) reaches production at peak capital intensity, and the US fab's share rises to 15-20%. The strategy should be to enjoy the "super window" certainty in 2026-2027 and adjust dynamically from 2028 based on the competitive landscape.
1. Company and products
- Founded by Morris Chang in 1987, pioneering the pure-play foundry model. 2025 global foundry share 69.9% (Counterpoint), controlling over 90% of sub-7nm advanced-process capacity and over 95% of AI-accelerator chip manufacturing.
- NVIDIA surpassed Apple as the largest single customer in 2025 at ~19% of revenue; the HPC platform was 61% of revenue in 2026Q1.
Revenue mix (FY2025 / 2026Q1)
| Dimension | Category | 2025 mix | 2026Q1 mix | 2025 YoY |
|---|
| Platform | HPC | 58% | 61% | +48% |
| Smartphone | 29% | 26% | +11% |
| IoT | 5% | 6% | +15% |
| Automotive | 5% | 4% | +34% |
| Process node | 3nm | 24% | 25% | — |
| 5nm | 36% | 36% | — |
| 7nm | 14% | 13% | — |
| Advanced total (≤7nm) | 74% | 74% | — |
| Region | North America | 75% | — | +42% |
| Mainland China | 9% | — | — |
2. Technology and packaging
Process roadmap (PPA comparison)
| Process | Baseline | Performance | Power | Density | Architecture | Production |
|---|
| N3E | vs N5 | +18% | -34% | 1.3x | FinFET | 2023 Q4 |
| N2 | vs N3E | +10-15% | -25~-30% | 1.15x | GAA nanosheet | 2025 Q4 |
| N2P | vs N3E | +18% | -36% | 1.20x | GAA | 2026 H2 |
| A16 | vs N2P | +8-10% | -15~-20% | 1.07-1.10x | GAA + SPR backside power | 2026-2027 H2 |
| A14 | vs N2 | +10-15% | -25~-30% | 1.20x+ | 2nd-gen GAA | 2028 |
- N2 entered production in Q4 2025 at Hsinchu Fab 20 and Kaohsiung Fab 22, SRAM yield >90%, early-2026 overall yield 60-70%, logic density 313 MTr/mm² (the industry's highest).
- 2nm monthly capacity rises from 30-50K wafers at end-2025 to 100-140K at end-2026; 2026 full-year capacity is sold out, with Apple taking >50% of early capacity.
Advanced packaging
- CoWoS capacity expands from ~35K wafers/month at end-2024 to a target 120-130K/month by end-2026 (~4x), still sold out through 2027. NVIDIA takes ~60-70%.
- Packaging revenue CAGR is expected >50%; packaging was ~8% of revenue in 2025, projected >10% in 2026.
Competitive comparison (2nm-class)
| Metric | TSMC N2 | Samsung SF2 | Intel 18A |
|---|
| Density (MTr/mm²) | 313 | ~231 | 238 |
| Yield (early 2026) | 60-70% | 50-60% | ~55% |
| Wafer ASP | ~$30,000 | ~$20,000 | undisclosed |
| Foundry share (2025) | 69.9% | 7.2% | <1% (external revenue $307M) |
3. Financials
| Metric | 2023 | 2024 | 2025 | 2026Q1 |
|---|
| Revenue (NT$B) | 2,161.7 | 2,894.3 | 3,809.1 | 1,134.1 |
| Revenue YoY | -4.5% | +33.9% | +31.6% | +35.1% |
| Net income (NT$B) | 838.5 | 1,173.3 | 1,697.6 | 572.5 |
| Gross margin | 54.4% | 56.1% | 59.9% | 66.2% |
| Operating margin | 42.6% | 45.7% | 50.8% | 58.1% |
| Net margin | 38.8% | 40.5% | 44.5% | 50.5% |
| ROE | 24.6% | 30.3% | 35.4% | 40.5% (annualized) |
| ROIC | 27.0% | 33.1% | 43.8% | — |
- 2026Q1 was the eighth straight quarter of double-digit profit growth; EPS NT$22.08 (5.7% beat). Q2 guidance: revenue $39-40.2B, gross margin 65.5-67.5%.
- Management raised full-year revenue-growth guidance from "~30%" to "over 30%" (USD), and the long-term gross-margin target from "53%+" to "56%+"; 3nm gross margin is expected to rise above the company average in H2 2026.
- 2026 capex guidance $52-56B (a record, 70-80% to advanced process, 10-20% to advanced packaging), fully self-funded by internal cash flow. 2025 FCF topped NT$1T for the first time at NT$1.003T.
- Dividend per share: 2025 NT$22.0, 2026E ≥NT$23 (+4.5%), payout ratio ~30%.
Valuation and peers (April 2026)
- Forward P/E ~27x, above the 5-year median of 22.1x, implying a ~3-5pp geopolitical discount; PEG ~0.6.
- Comparison: NVIDIA ~35x, ASML 36x, Samsung 18.2x. TSMC's "faster growth, lower valuation" reflects the Taiwan-Strait risk premium (Berkshire's ~60% cut in 2024Q4 was based on this).
4. Sentiment and institutions
- Of 37 analysts, 30 Buy / 6 Hold / 1 Sell, an 81% buy consensus; ADR consensus target ~$215 (Morgan Stanley 210, Goldman 218, JPM 215, Bernstein 225, UBS 205).
- Passive funds (Vanguard, BlackRock) and Taiwan domestic capital (25-30% index weight) provide a floor; active funds (Berkshire, ARK) cut on geopolitics/rotation.
5. Competitive landscape
- Samsung Foundry holds 7.2% share (still loss-making, 2025 revenue -3.9%), 2nm yield recovered to 50-60%, targeting profit and 20% share in 2027 (unprecedented); SF1.4 pushed to 2029.
- Intel 18A reached production in 2026Q1 at ~55% yield, with external revenue of just $307M; Amazon Trainium3 chose 18A (the first major cloud-provider AI chip to leave TSMC), and it participates in Musk's Terafab.
- Mainland China is aggressively expanding mature nodes, possibly causing 28nm+ oversupply after 2027, with TSMC's mature-node gross margin potentially falling from 65-75% to 40-50% (mature process is now ~26% of revenue).
6. Investment potential (short/mid/long term)
- Short term (2026-2027 super window): AI demand explosion + 2nm pricing power + CoWoS sold out = very high earnings certainty. AI-accelerator revenue was ~$33.4B in 2025 (+100%+), and management raised the 2024-2029 AI-accelerator revenue CAGR from "mid-40%" to "mid-to-high 50%." Short-term target ADR $210-225 (25x 2026E EPS).
- Mid term (2027-2028): A16 production brings a new growth engine, but the US-fab ramp dilutes gross margin 2-3pp/yr (later 3-4pp), and with 2nm-ramp dilution the post-2027 pressure could reach 5-7pp, with overall gross margin falling to 56-60%.
- Long term (2028-2030): ASML EUV equipment annual-output bottleneck (70-80 units) limits both rivals' catch-up and TSMC's own expansion; Taiwan-Strait geopolitics is a tail risk (Polymarket 2026 conflict odds 16%).
Mid-term scenarios (EPS, ADR basis)
| Scenario | 2027E EPS | 2028E EPS | Probability | Core assumptions |
|---|
| Bull | $21.50 | $25.00 | 25-30% | A16 smooth, CoWoS tight, gross margin >60% |
| Base | $18.05 | $20.50 | 50-60% | A16 normal ramp, supply-demand balance, gross margin 56-58% |
| Bear | $14.50 | $15.00 | 15-20% | AI capex slows, CoWoS surplus, gross margin 52% |
7. Strategy and verification metrics
- Position: 15-25% of a tech portfolio, single stock no more than 10% of the portfolio. Overweight in 2026-2027, adjust from 2028 by the competitive landscape.
- Entry: a pullback below 22x forward P/E (ADR $160-180) is the ideal add point; stop at ADR $150.
Key verification metrics (excerpt)
| Dimension | Metric | Positive signal | Negative signal |
|---|
| Financial | Quarterly gross margin | >58% and stable/rising QoQ | <55% for two quarters or QoQ drop >2pp |
| Financial | EPS growth (YoY) | >30% | <15% for two quarters |
| Tech | 2nm yield | 2027 >75% | <55% |
| Tech | CoWoS capacity utilization | >90% | <75% |
| Market | NVIDIA order growth | YoY >25% | YoY <0% or two quarters of negative QoQ |
| Geo | US-fab gross margin | improves from 8% to 15%+ | persistently <10% with no improvement |
| Geo | Taiwan-Strait risk (Polymarket) | <15% | >25% or military escalation |
Exit signals (priority)
- Gross margin declines two straight quarters → immediately trim 50%
- NVIDIA order growth turns negative → cut to neutral weight within 1 month
- Forward P/E >35x (not a geopolitical-easing case) → cut to neutral weight
- Geopolitical risk escalates sharply (Polymarket >30% or an actual military event) → exit immediately
8. Risks and conclusion
Main risks
| Risk | Alert | Impact | Probability | Potential damage |
|---|
| Taiwan-Strait conflict/blockade | Red | Very high | Medium | Revenue disruption, 30-50% valuation loss |
| US export-control escalation | Red | High | Medium-high | China-customer revenue to zero (~10-15%) |
| CHIPS Act conditions | Red | Medium-high | Medium-high | Gross-margin dilution 2-4pp |
| AI-training-investment overheating pullback | Orange | Medium-high | Medium-high | HPC revenue down 15-25% |
| Samsung 2nm yield catch-up | Yellow | Medium-high | Medium | Weaker pricing power, gross margin -1-2pp |
| China mature-node expansion | Yellow | Medium | High | Mature-node gross margin -15-25pp |
| US-fab cost overruns / EUV bottleneck | Operational | Medium | Medium-high | Extra capex $20-50B / capacity delays 1-2 years |
Final rating
Buy / Core Holding; 12-month target NT$2,100 / ADR ~$220 (2026E EPS NT$80-85 × 25x). This rating assumes 2026-2027 fundamental certainty continues and geopolitics does not deteriorate to an extreme; re-evaluate and dynamically adjust the position by the competitive landscape in late 2027 to early 2028.
Note: the original report is internally inconsistent on the absolute ADR price level (the financial deep-dive section cites ADR ~$402, market cap $1.79T, EPS $15.45, P/E 27x; while the strategy/conclusion sections use ADR ~$185 as the "current price" and $220 as the target). This structured data uses the strategy/conclusion basis consistent with the "Buy" rating and target (current ADR ~$185, target ~$215-220).