§ 01Executive View
TSMC has the strongest customer base in the entire semiconductor industry: every fabless leader at the leading edge is a customer, switching costs are measured in years and billions of dollars, and the customer mix is migrating to the highest-quality demand pool on Earth (HPC/AI). Concentration is real — Apple alone is roughly a quarter of revenue and the top ten customers cluster well above 70% — but the concentration is into the winners, not into a single fragile account, and HPC has now overtaken smartphone as the largest end-market, structurally lengthening the demand cycle. The customer dimension supports a high-conviction long: TSMC's customers are fighting each other for capacity, not negotiating it down.
§ 02Customer Concentration
| Metric | Latest | YoY change | Source |
|---|---|---|---|
| Top customer % (Apple, est.) | ~23–25% | Flat-to-down (HPC mix dilution) | TSMC 20-F segment color; supply-chain teardowns; estimated |
| Top 3 (Apple + NVIDIA + AMD/Qualcomm) | ~50–55% | Up — NVIDIA now #2, displacing prior smartphone customers | Industry estimate; TSMC does not name customers |
| Top 5 | ~65% | Up | Industry estimate |
| Top 10 | ~75–80% | Roughly flat | Industry estimate |
| Named >10% customers in 20-F | One unnamed customer disclosed at >10% (Apple, by market consensus) | Stable | TSMC 20-F |
TSMC does not disclose customer names; the >10% disclosure in the 20-F is the regulator-compelled minimum. Synthesis cohort note flags Apple at ~25% historically with NVIDIA rapidly rising — likely already #2 and potentially closing the gap on Apple in absolute dollars as Blackwell + Rubin + CoWoS allocations ramp through 2026. The list of routine top-10 occupants is now: Apple, NVIDIA, AMD, Qualcomm, MediaTek, Broadcom, Marvell, Sony, Intel (as foundry customer), plus the hyperscaler-custom-silicon designs (Google TPU, AWS Trainium, Microsoft Maia, Meta MTIA, OpenAI/Broadcom 2027 chip) — most of which are routed through Broadcom or Marvell as design partners but ultimately consume TSMC wafers and CoWoS capacity. The trend in concentration is up at the node level (N3/N2/A16 are even more concentrated among ~6 customers) but down in dispersion-of-end-customers terms because hyperscaler captive silicon adds 4–5 new top-20 buyers.
§ 03End-Market Exposure
| Segment | % rev (2024–25 est.) | Cycle position | Structural | Macro sensitivity |
|---|---|---|---|---|
| HPC / AI | ~50–55% (now #1) | Early-to-mid expansion | Secular growth — every model generation pulls more wafers | Moderate; capex-funded by hyperscalers with $600B+ 2026 budget |
| Smartphone | ~28–30% | Late cycle; mature | Flat-to-low-growth in units; ASP-led via N3/N2 at Apple | High to consumer cycle |
| IoT | ~5% | Trough | Structurally slow | Moderate |
| Automotive | ~5% | Mid-correction (digesting inventory) | Long-term growth (more silicon per car); near-term soft | High to auto cycle |
| DCE (data center / consumer electronics, residual) | ~7–10% | Mixed | Mixed | Moderate |
The single most important fact in the table is that HPC overtook smartphone in 2024–25 and is now the largest segment. Smartphone was historically the highest-volatility part of TSMC's mix, with the September quarter "iPhone build" cycle dominating the print. Replacing that with hyperscaler AI demand — which is buy-the-capacity-you-can-get rather than negotiate-on-price — meaningfully lengthens the demand cycle and raises the floor on utilization. Auto softness is real but at ~5% of mix it cannot move the company; it is more relevant as a free option on the auto cycle eventually re-accelerating into the 800V/SiC ecosystem the cohort synthesis flags as a parallel demand curve. The weighted demand picture is: HPC compounding at hyperscaler-capex pace (multi-hundred-billion annual), smartphone steady, everything else marginal. This is the most favorable end-market mix in TSMC's history.
§ 04Contract Structure & Switching
Switching cost is the highest in the entire semiconductor stack. A new chip designed on TSMC's N3 or N2 PDK requires ~18–36 months of tape-out, qualification, and yield-tuning work specific to TSMC's design rules, IP libraries (ARM, Synopsys, Cadence on TSMC PDK), and back-end (CoWoS-L, SoIC). Porting to Samsung Foundry or Intel Foundry is not a procurement decision; it is a re-architecture project costing tens to hundreds of millions per chip and delaying time-to-market by a year or more. NVIDIA has explored Samsung at the margins; the volume parts (Blackwell, Rubin) are TSMC-only. Apple has never moved off TSMC since A8.
Contract structure is multi-modal:
- Apple pre-payment model: Apple pre-pays hundreds of millions per node in exchange for first-dibs capacity; per the cohort synthesis "without Apple's pre-payment, TSMC's capital cycle wouldn't work." This is the prototype customer-funded capex model.
- NVIDIA / hyperscaler advance reservations: Multi-quarter capacity reservations at N3/N2/A16 plus CoWoS-L allocation slots — effectively take-or-pay because the reservation deposits are non-refundable and customers compete for them.
- Long-term agreements for foundation customers (Apple, NVIDIA, AMD, Broadcom) span multi-year tape-out roadmaps tied to specific node ramps.
- Spot / merchant exposure is a small minority and concentrated in trailing-edge nodes (N16/N28) where switching cost is lower.
Forward visibility is strong but not formally disclosed as RPO: TSMC guides via revenue outlook and HPC mix commentary on the quarterly call, and the CoWoS waitlist itself is a leading indicator — currently sold out through 2026 with capacity being doubled to 80k+ wafers/mo by end-2026 per the cohort catalyst list. Customer concentration is widening at the top by adding hyperscaler captive silicon programs (Google TPUv7, Trainium3, Maia 2, Meta MTIA, OpenAI's 2027 Broadcom chip) — every one of those programs is a new TSMC customer that didn't exist five years ago.
§ 05Demand Quality
This is where TSMC's customer-side story is structurally cleanest. The HPC/AI demand currently flowing is overwhelmingly pull-through, with very limited evidence of channel fill or pre-buy distortion:
- Pull-through: Hyperscaler 2026 capex of ~$600B and ~50 GW of new AI capacity (per cohort synthesis) is matched against actual deployment at Stargate, Colossus, Project Rainier, Anthropic's TPUv7 commitment, etc. These wafers are being installed in real datacenters with real PPAs and real interconnect queues.
- Channel fill risk is structurally low because TSMC sells to fabless designers, not to distributors. Inventory cycles do exist (Apple iPhone builds, smartphone SoC pre-loads) but those are at the smartphone end of the mix and have historically been transparent on the call.
- Pre-buy is partially present in CoWoS-L reservations — customers are over-ordering capacity slots given uncertainty on AI ramp pace — but this is a positive form of pre-buy because it commits non-refundable deposits and, if anything, anchors revenue rather than threatening it.
- Watch item: smartphone N3 utilization in the back half of 2026 if iPhone units disappoint. This is the only soft spot.
Backlog burn rate: CoWoS slots are burning at the rate TSMC can install tools — capacity-constrained, not demand-constrained. The cohort note that "Nvidia is supply-constrained" with the constraint being "almost always CoWoS-L capacity at TSMC, not chip dies" is the cleanest possible signal of demand quality — customers are pulling, not being pushed.
§ 06Bull Points
- HPC overtook smartphone as #1 segment in 2024–25, structurally lengthening the demand cycle and replacing the most volatile part of historical revenue.
- Customer list is the entire merchant leading-edge fabless industry — Apple, NVIDIA, AMD, Qualcomm, MediaTek, Broadcom, Marvell, Sony, Tesla, plus every hyperscaler captive-silicon program (TPU, Trainium, Maia, MTIA, OpenAI). New top-20 customers are being added, not lost.
- Switching cost is multi-year and multi-hundred-million-dollar per chip program — Samsung Foundry yield problems and Intel Foundry ecosystem gaps mean there is effectively no second source for N3/N2/A16 + CoWoS-L.
- Apple pre-payment + NVIDIA reservation deposits = customer-funded capex, an unusual model where the customer base finances the supplier's growth.
- CoWoS-L is sold out through 2026 with non-refundable allocation slots; this is the strongest forward visibility in the sector.
§ 07Bear Points
- Apple at ~25% is real concentration: an iPhone unit shock or a hypothetical Apple-Samsung detente would create a meaningful one-quarter air pocket, even if structurally Apple has nowhere else to go.
- NVIDIA rising to potentially #2 at ~15–20% creates a second concentration vector — a Blackwell yield issue or Rubin delay would directly compress TSMC HPC revenue.
- Hyperscaler custom silicon is a double-edged sword: every TPUv7 sold is great for TSMC (the wafer comes from them anyway) but it caps NVIDIA pricing power, and if NVIDIA ASPs eventually compress, NVIDIA's CoWoS allocation budget per dollar of revenue may shrink.
- Auto + IoT = ~10% combined and both in correction; not a thesis-mover but a 1–2% drag.
- Taiwan tail risk is the customer-side analog of supply concentration — a Taiwan disruption would not just halt TSMC's production, it would also halt its customers' ability to ship anything leading-edge for 18+ months.
- Top-customer concentration drift is slightly upward at the node level: N2 and A16 are designed in by ~6 customers, not 60. If any one of them defects or stumbles, the impact on bleeding-edge revenue is non-trivial.
§ 08Conviction (1–5)
5. Customer concentration is genuinely high but it is concentrated into the strongest counterparties in technology, the demand mix has migrated to the highest-quality end-market (HPC/AI), switching costs are extreme, contracts are long-term and partly customer-pre-funded, and demand is overwhelmingly pull-through with CoWoS sold out. There is no obvious customer-side deterioration scenario short of a Taiwan geopolitical event, which is a supply-side / existential issue rather than a customer issue.
§ 09Key Risks to This Read
- Apple iPhone unit disappointment in late 2026 / 2027 would expose the smartphone tail of the mix.
- NVIDIA pricing power compression from custom silicon could indirectly reduce NVIDIA's per-wafer profitability and pressure CoWoS allocation budgets — though TSMC gets paid either way (Broadcom/hyperscaler ASIC wafers are TSMC wafers too).
- Customer movement to Intel 18A / 14A or Samsung SF2 if either fab solves yield — currently very unlikely in the 2026–2027 window per the cohort synthesis but not zero.
- Hyperscaler capex air-pocket — if 2027 capex guides flat after $600B in 2026, HPC growth could decelerate; demand quality concern though not a destruction-of-customer-base concern.
- Auto cycle persistence at trough — small mix item but if it stays soft for 4+ quarters, removes a positive surprise vector.
§ 10Sources
- TSMC entry in cohort
companies.json(id 2): "TSMC is irreplaceable… Nvidia is supply-constrained, the constraint is almost always CoWoS-L capacity at TSMC, not chip dies." - Cohort synthesis.md Sections 2 (value chain), 3.4 (three bottlenecks), 3.7 (custom silicon), 5 (tailwinds/headwinds), 7 (company universe).
- TSMC 20-F (most recent) — customer concentration disclosures (one unnamed >10% customer, segment revenue mix, geographic mix).
- TSMC quarterly earnings calls Q3'24–Q1'26 — HPC overtaking smartphone commentary, CoWoS capacity-doubling guidance, end-market color.
- Industry supply-chain analyst estimates (cohort cross-references) for top-10 customer composition (Apple, NVIDIA, AMD, Qualcomm, MediaTek, Broadcom, Marvell, Sony, Intel, hyperscalers via Broadcom/Marvell).
- User notes provided in cohort context: Apple ~25%, HPC > smartphone, hyperscaler custom-silicon list, switching-cost framing.
Works cited
- TSMC Q4 2025 Earnings Call Transcript
- Q4 2025 GM 62.3% above 59-61% guide; FY2025 GM 59.9% +380bps YoY; FY2026 GM guide 63-65%; pricing 'strategic, not opportunistic'
- Counterpoint Global Pure Foundry Market Share Quarterly
- Quarterly share series TSMC/Samsung/SMIC; Intel Foundry 6% in Foundry 2.0 frame
- Samsung 2nm Yields ~55%, Below MP Threshold (TrendForce, Apr 2026)
- Samsung 2nm yield ~55% as of April 2026, below ~60% MP threshold; Qualcomm leaning back to TSMC
- Samsung Lands $17B Tesla AI6 Foundry Deal (TrendForce)
- Tesla AI6 $16.5-17B Samsung Foundry win at Taylor TX; first major external 2nm-class commitment for Samsung in years
- Samsung vs TSMC vs Intel Foundry Market Numbers (PatentPC)
- Q3 2024 baseline TSMC 64.9% / Samsung 9.3%; Samsung dual-role IDM/foundry conflict; customer-flight pattern
- TrendForce 2Q25 Foundry Revenue 14.6% Up, TSMC 70%
- TSMC Q2 2025 share 70.2%, revenue $30.24B
- TSMC 2nm 50K to 140K Wafers in 2026 Supply Shock (StreetStocker)
- N2 capacity ramp 40k -> 100k wafer/mo 2026, 200k by 2027; demand exceeds initial ramp
- TSMC 2nm Up 10-20%, 3-7nm Single-Digit Hikes 2026 (TrendForce)
- N2 wafer ~$30k +10-20% above N3; N3-N7 single-digit hikes 2026
- TSMC CoWoS-L/S Fully Booked, OSAT Partners Step Up (TrendForce)
- CoWoS-L/S sold out; ASE CoWoP / Amkor stepping up as overflow OSAT alternatives
- TSMC Q4 FY 2025 Results and FY 2026 Outlook (Futurum)
- Q4 2025 GM and FY2026 guidance commentary; 56%+ long-term GM target reaffirmed
- TSMC Samsung Intel Who's Leading the Semiconductor Race (PatentPC)
- Qualcomm 8 Gen 1 35% Samsung yield vs 70% TSMC 4nm; subsequent migration to TSMC for all flagship Snapdragons
- China to Increase Leading-Edge Output 5x in Two Years (Tom's Hardware)
- China 7nm/5nm capacity targets - 100k wafer/mo by 2027-28, 500k by 2030; SMIC 7nm yield 60-70%
- Intel CEO Embraces 18A for External Customers (Tom's Hardware)
- Intel 18A external engagement; CFO acknowledged committed external 18A volume 'not significant' as of mid-2025
- Intel Foundry Reportedly Secures 18A for Microsoft Maia 3 (TechPowerUp)
- Microsoft Maia 2/3 anchor commitment to Intel 18A/18A-P
- Intel Going Big Time Into 14A - Lip-Bu Tan (Tom's Hardware)
- Intel 14A PDK distribution; two test-chip evaluators on Q4 2025 call; zero firm 14A external commitments
- NVIDIA Alone Has TSMC Advanced Packaging Booked Years Ahead
- NVIDIA wafer / advanced-packaging book through 2027
- Rapidus Lands $1.7B to Chase 2nm by 2027 (The Register)
- Rapidus $1.7B Feb 2026 tranche; IBM tech transfer; Tenstorrent first announced customer; 2nm risk production target 2027
- Samsung Hits 70% Yield on 2nm GAA SF2P (FinancialContent, Jan 2026)
- Samsung SF2P 70% yield headline (contradicted by April 2026 reporting); first credible 2nm second-source signal
- SMIC On Track to Produce 5nm for Huawei (Tom's Hardware)
- SMIC 5nm pilot 2026 for Huawei Ascend / Alibaba; DUV-only constraint
- TSMC Boosts CoWoS, NVIDIA Dominates Advanced Packaging Through 2027
- CoWoS scaling 35k -> 130k wafer/mo by end-2026; NVIDIA >50% allocation through 2027; 510k CoWoS-L wafers booked for Rubin/Vera/GB100
- TSMC Nears 70% Foundry Share, Gap with Samsung 62.7 pp (BigGo)
- TSMC ~70% pure-play foundry share 2025; gap to Samsung widened to 62.7 pp
- TSMC Q1 2026 Revenue and 66.2% Gross Margin
- Q1 2026 GM 66.2% above 63-65% guide; demonstrates active pricing power
- TSMC to Raise Advanced Node Quotes Up to 10% in 2026 (Tom's Hardware)
- 5-10% sub-5nm hikes 2026; Arizona ~15% premium, N5/N4 25% premium uniformly applied
- TSMC to Raise Prices for Four Consecutive Years From 2026 (WCCFTech)
- Customers notified of 4-year consecutive price hike cycle on advanced nodes
- Why TSMC Grew 4x Faster Than Foundry Rivals in 2025 (Tom's Hardware)
- TSMC growth rate vs rivals; price hikes + vertical integration + technology lead synthesis
- 24/7 Wall St - AI Demand Has Permanently Rewired Semiconductor Pricing
- Hyperscaler long-dated supply commitments dampening foundry inventory cycle
- Astute Group - Advanced Packaging Demand Soars: Nvidia Secures 60% of CoWoS
- Advanced packaging market sizing $49-55B by 2026
- CoWoS allocation
- BIS December 2, 2024 HBM Rule
- HBM density caps for PRC consumption
- bears on TSMC base-die packaging for restricted parties
- BIS Entity List actions and license-suspension notices re: Sophgo (late 2024)
- Underlying enforcement actions tied to Sophgo/Huawei Bingchuan-chip incident
- BIS October 17, 2023 export control rule update
- Expanded thresholds and entity scope
- further constrains TSMC PRC AI book
- BIS October 7, 2022 export control rule (advanced computing and SME to PRC)
- Foundation of FDPR sub-7nm restrictions binding TSMC for PRC fabless customers
- BIS press release — Commerce Strengthens Export Controls (Dec 2024)
- Dec 2024 SME / HBM rule scope
- BIS press release — Foreign-Owned Fab Loophole Closed
- Foreign-owned fab perimeter expansion
- Cohort companies.json (TSM entry id=2)
- Supporting quotes, catalysts (N2 ramp, A16, CoWoS doubling), risks (Taiwan geopolitical concentration, Arizona ~30% cost premium, capex/transistor flatlining)
- cohort companies.json — TSM entry (id 2)
- Customer concentration framing
- CoWoS as binding constraint quote
- catalysts and risks
- Cohort synthesis (chokepoint thesis, hyperscaler $600B capex, CoWoS bottleneck)
- Reverse DCF anchor: AI-cycle structural growth assumptions, CoWoS-L as binding constraint, Arizona 30% cost premium framing
- Cohort synthesis and ledger
- Context on export-control framing and Taiwan tail risk per user notes
- Cohort synthesis.md
- TSM macro positioning, Taiwan tail framing as 'existential', three-bottleneck thesis, AI capex aggregate (~$600B / 50 GW), FX cohort context, cyclicality framing
- cohort synthesis.md — Sections 2, 3.4, 3.7, 5, 7
- Value chain map, three-bottleneck framing, custom-silicon dynamics, hyperscaler $600B capex, end-market context
- Commerce CHIPS Program Office — Preliminary Memorandum of Terms with TSMC Arizona
- $6.6B direct funding
- capacity covenants
- clawback and PRC-expansion guardrails
- Coordination handoff with regulatory-analyst (BIS rule mechanics, CHIPS Act, FDPR)
- Macro lane covers trade-flow direction and FX consequences
- regulatory lane owns specific rule mechanics. Avoid double-counting per contract.
- Counterpoint Research - Global Pure Foundry Market Share Quarterly
- Pure-play foundry quarterly share by player
- Q3'25 prints (TSMC ~71%, Samsung ~6.8%)
- HHI inputs
- Covington & Burling — US Strengthens Export Controls on Advanced Computing and SME (Dec 2024)
- Legal-analysis perspective on Dec 2024 rules
- CRS — U.S. Export Controls and China: Advanced Semiconductors (Aug 2025)
- Export-control regime trajectory and TSMC China-customer revenue exposure
- Deloitte Insights - 2026 Semiconductor Industry Outlook
- $975B-$1T market 2026 sizing
- 26% YoY growth
- Digitimes - TSMC unveils four-year price hike for advanced chips starting 2026
- Four-consecutive-year ASP runway 2026-2029
- ASP discipline thesis
- Entropy Capital — ASML's Supply Chain, Bill of Materials
- ASML BOM and tier-2 dependency map
- Epoch AI — Advanced packaging and HBM, not logic dies, were the bottlenecks on AI chip production in 2025
- CoWoS-as-binding-constraint corroboration
- EU Commission State aid Decision SA.107536 — Germany — ESMC Dresden
- EUR 5.0B state-aid clearance for ESMC JV
- capacity and operations conditions
- Fabricated Knowledge (Doug O'Laughlin) - 2026 AI & Semiconductor Outlook
- Cycle framing - mid-supercycle, GB200 inventory partially resolved, expansion phase
- FinancialContent - High-Stakes Gamble: Intel Foundry Resurgence vs TSMC 2026
- Intel Foundry Q3'25 revenue ($223M, ~0.5%)
- Nvidia 18A pause
- competitive depth
- FinancialContent - The Great Packaging Pivot: TSMC Doubling CoWoS Capacity
- CoWoS allocation 2026 (Nvidia 60%, Broadcom 15%, AMD 11%, >85% pre-allocated)
- FinancialContent - The Silicon Mosaic: Chiplets and the UCIe Standard
- Disruption-watch on UCIe maturation
- 120+ consortium members
- mainstream 2026 adoption
- Fortune Business Insights - Semiconductor Foundry Market Forecast [2034]
- TAM 2026 estimate ($202B) - upper bound on consensus range
- GII Research / KSI - Semiconductor Foundry Market Forecasts 2025-2030
- 5y CAGR triangulation
- 10/7/5nm-and-below tier at 28.3% CAGR
- Global Market Insights - Semiconductor Foundry Market Size Growth Report 2035
- TAM 2026 estimate (~$180B)
- 5.6% 2024-2030 CAGR data point
- GuruFocus — TSM EV/EBITDA Historical (10y range)
- 10-year EV/EBITDA range (5.45 min / 9.26 median / 19.24 max) for historical context
- IDC - Semiconductor Foundry 2.0 Market Entering Growth Phase from Recovery (11% YoY 2025)
- Foundry 2.0 sizing
- recovery-to-expansion cycle phase
- TSMC 37% Foundry 2.0 share
- Industry supply-chain analyst estimates (TrendForce, SemiAnalysis, DIGITIMES) — cohort cross-references
- Top-10 customer composition estimate (Apple, NVIDIA, AMD, Qualcomm, MediaTek, Broadcom, Marvell, Sony, Intel, hyperscaler ASICs)
- estimated concentration percentages — flagged as estimates
- METI JASM Phase 1 (Dec 2021) and Phase 2 (Dec 2023) subsidy decisions
- JPY 476B and JPY 732B subsidy commitments to TSMC Kumamoto fabs
- Mordor Intelligence - Semiconductor Foundry Market Analysis
- TAM 2026 estimate ($184.78B)
- 7.42% CAGR 2025-2030
- Multiples.vc — TSMC Public Comps
- Cross-sectional valuation comparison vs WFE chokepoints (ASML, AMAT, KLAC) and foundry peers (UMC, GFS)
- PatentPC - TSMC, Samsung, Intel: Who's Leading the Semiconductor Race
- Leading-edge share at 3nm/2nm (TSMC 90%+)
- competitive positioning vs Samsung/Intel
- Public defense and strategic analyst commentary on Taiwan Strait scenarios
- Probability framing for blockade vs kinetic scenarios
- mechanism (PLA exclusion zone, fuel-reserve depletion, cable-cuts gray-zone). Probability bands are analyst judgment.
- Public macroeconomic regime references (FRED US 10y, DXY, USD/TWD spot, JPY/USD)
- Current regime: USD strength, US 10y 4-4.75% range, TWD weakness 2024-2026, JPY weakness 2022-2026
- Section 232 semiconductor investigation initiation notice
- Live investigation, statutory 270-day clock, presidential decision window
- Section 48D Advanced Manufacturing Investment Tax Credit — final regulations
- 25% ITC structure for TSMC Arizona qualified property
- SemiWiki - CoWoS Capacity Set to Skyrocket by 2026
- CoWoS capacity scaling 35K -> 130K WPM by EOY26
- 1M wafers demand by 2026
- Semiwiki - TSMC 2025 Update: Riding the AI Wave
- TSMC 2025 revenue (~$122.5B, +36% YoY)
- operational scale data
- SpecGas — Neon Production by Country 2026
- Post-2022 neon diversification (China-led, US/Korea capacity additions)
- StockAnalysis.com — GFS, UMC statistics pages
- GlobalFoundries (EV/EBITDA ~10.7x, fwd P/E ~22x) and UMC (EV/EBITDA ~5.5x, fwd P/E ~15x) for relative valuation
- StockAnalysis.com — TSM income statement / cash flow / balance sheet / statistics
- FY22-FY25 historical income / cash flow / balance sheet series
- current multiples (P/E 30.6x, fwd P/E 20.5x, EV/EBITDA 18.8x, EV/Sales 13.1x, FCF yield 1.9%)
- ROIC 52%, ROE 36%
- Taipower historical industrial rationing episodes (notably 2021 drought)
- Taiwan power and water as quasi-macro operating risk
- energy-import dependency context (~97% imported)
- fuel-reserve duration (~40 days)
- Taiwan Statute for Industrial Innovation Article 10-2 (Taiwan CHIPS Act)
- 25% R&D tax credit plus 5% advanced equipment credit
- effective-tax-rate floor
- TechSoda — Explainer: TSMC's 2024 Annual Report Highlights
- Annual report supplier-list synthesis
- Tom's Hardware - Semiconductor industry enters unprecedented giga cycle
- Giga-cycle structural framing of 2026 industry dynamics
- TrendForce - TSMC 2nm 60K Monthly Output 2026, Prices 50% Above 3nm
- N2 pricing premium (~50% above N3)
- 2nm capacity 60K WPM 2026
- TrendForce - TSMC 2nm Reportedly Up 10-20%; 3-7nm Single-Digit in 2026
- 2026 wafer pricing (N2 +10-20% over N3
- N3-N7 single-digit increases)
- Trendforce — ASML's Magic Uncovered: Tech and Partners Behind Its EUV Edge (Nov 2025)
- ZEISS / Cymer / TRUMPF tier-2 dependency mapping for ASML EUV
- Trendforce — Japan Ramps Up Photoresist Investment for 2nm Chips (Nov 2025)
- Photoresist supplier concentration, TOK Korea plant capex
- Trendforce — Japan Rumored to Curb Photoresist Exports (Dec 2025)
- Photoresist export-control geopolitical context
- Trendforce — Kioxia, TEL and Photoresist Makers in Focus After M7.7 Japan Earthquake (Apr 2026)
- April 2026 photoresist disruption stress-test data point
- Trendforce — TSMC Accelerates Arizona 2nd Fab, Eyes 3Q26 Tool Install (Dec 2025)
- Arizona Fab 2 timeline acceleration
- Trendforce — TSMC Reportedly Plans 12 New Advanced Process and Packaging Fabs in Taiwan (Nov 2025)
- Taiwan capacity concentration vs ex-Taiwan ramp
- Trendforce — TSMC Reportedly Pulls Arizona Third Fab to 2027 (Sep 2025)
- Geographic diversification ramp acceleration
- TSMC 1Q26 Management Report
- Q1 2026 P&L, segment mix (HPC 58%, smartphone 29%, advanced nodes 74% of wafer revenue), capex, cash balance
- TSMC 2024 Annual Report
- Tier-1 supplier disclosures, capex allocation, EUV scanner counts
- TSMC 2024 Responsible Supply Chain Report
- Continuity planning posture, 2,000+ chemical/material qualification, neon recycling
- TSMC 2025 Annual Report (English)
- FY25 segment mix, capital allocation framework, dividend policy, capex history
- TSMC 2025 SEC 20-F
- FY25 annual report regulatory filing for ADR
- cash flow detail
- SBC disclosure
- TSMC 4Q25 Earnings Call Transcript
- Capacity utilization, Arizona ramp commentary, capex guidance
- TSMC 4Q25 Management Report
- FY25 full-year P&L, capex (~NT$1.27T / ~$40B), balance sheet
- TSMC Arizona corporate site
- Arizona Fab 1 N4 HVM, yield parity statements
- TSMC Form 20-F (most recent annual filing)
- Customer concentration disclosure (one unnamed >10% customer), geographic/end-market revenue mix
- TSMC Form 20-F (most recent, FY2024)
- Risk factors, government grants disclosure, material litigation note
- TSMC FY24 Annual Report and 20-F filings (general reference)
- FX sensitivity rule-of-thumb (~40 bps GM per 1% TWD/USD move), revenue mix by geography, capex profile, segment mix HPC/smartphone/auto/IoT, balance sheet net cash position
- TSMC IR — Q1 2026 Quarterly Results page
- Q1 2026 revenue, gross/operating/net margins, capex, Q2 and full-year 2026 guidance
- TSMC Q1 2026 Earnings Call Transcript (Investing.com)
- Q1 2026 OCF (NT$699B), FCF (NT$348B), ROE (40.5%), raised long-term GM target (56%+), 2026 capex high-end of $52-56B range
- TSMC quarterly earnings calls (Q3 2024 through Q1 2026)
- HPC overtaking smartphone commentary, CoWoS capacity-doubling guidance, end-market segment color, demand-quality signaling
- TSMC quarterly earnings transcripts FY24-Q1 FY26 (general reference)
- Wafer pricing direction (2023 6-8% raise, 2024 ~3%, N2/A16 reported ~10-15% premium), China revenue trajectory, Arizona ramp commentary, water/power risk mentions
- User-provided cohort context for customer dimension
- Apple ~25%, HPC overtook smartphone in 2024–25, hyperscaler custom-silicon list, switching-cost description, cycle-position read by end-market
- USITC — Ukraine, Neon, and Semiconductors (executive briefing)
- Pre-2022 neon supply baseline
- Wccftech - TSMC Tight 2nm Supply, Four Consecutive Years of Price Hikes
- Confirms multi-year price-hike runway
- supply tightness at N2
- Wccftech — TSMC 2nm Tight Supply / Four-Consecutive-Year Price Increases
- Pass-through pricing power evidence
- Works in Progress Magazine — The world's most complex machine (ASML/EUV)
- ZEISS / Cymer / TRUMPF technical context