Taiwan Semiconductor Manufacturing Company Limited

TSMC — Business Economics

Ticker: TSM | Currency: USD (ADR on NYSE; underlying shares trade in TWD on TWSE)

TSMC is the world's most important semiconductor company and the purest expression of a structurally advantaged business model in technology. It makes money by manufacturing chips designed by others — a pure-play foundry. Its economic engine is accelerating, not decelerating.

The DNA. TSMC does not design chips. It builds them under contract for ~500+ customers including Apple, NVIDIA, AMD, Qualcomm, and Broadcom. Revenue is a function of wafer volume × technology-node pricing, with leading-edge nodes (3nm, 5nm) commanding substantial premiums. In FY2024, revenue reached NT$2,894B (~US$88.5B), up 34% YoY, with gross margins of ~59% and operating margins approaching 48% — extraordinary for a capex-heavy manufacturer. HPC (AI/datacenter) overtook smartphones as the dominant revenue segment at ~51%, with smartphones at ~35%.

Where it's headed. The business is inflecting upward, not plateauing. Three forces drive this: (1) AI training and inference demand is pulling massive GPU/accelerator volumes through TSMC's most advanced nodes and advanced packaging (CoWoS); (2) leading-edge concentration is intensifying — TSMC's 3nm/5nm nodes accounted for ~69% of wafer revenue in FY2024, up from ~54% in 2023; (3) competitors cannot close the gap — Intel Foundry and Samsung Foundry continue to lose share and yield ground. FY2025 revenue (per the 20-F filed with the SEC for the year ended December 31, 2025) extended this trajectory with continued AI-driven growth.

Win-win model. TSMC's foundry model is genuinely symbiotic. Fabless designers (NVIDIA, AMD, Apple) get world-class manufacturing without $20B+ fab investments. TSMC gets diversified demand across hundreds of customers. No single customer except Apple exceeds ~25% of revenue, and even that dependency runs both ways — Apple cannot leave.

Key metrics to track: (1) advanced-node revenue share (higher = stronger pricing power), (2) gross margin trajectory (proxy for technology leadership and pricing), (3) capex-to-revenue ratio (capacity investment vs. returns), (4) HPC revenue mix (proxy for AI exposure), (5) capacity utilization rates.

Signs of deterioration? None. The opposite — TSMC is experiencing the most favorable demand environment in its history, with structural AI tailwinds layered on top of its existing smartphone and HPC franchise.

Market Overview — TSMC

TSMC operates in the global semiconductor foundry market, which has evolved from a commoditized wafer-baking service into the single most critical bottleneck in the technology supply chain. The foundry TAM reached roughly $145–150 billion in 2025 and is projected to grow at a mid-teens CAGR through 2030, propelled by AI accelerator demand (GPUs, custom ASICs), HPC, and the proliferation of silicon into automotive and edge devices. AI-related revenue alone grew from ~mid-teens percent of TSMC's revenue in 2024 to over 20% in 2025, and is the fastest-expanding segment.

Competitive landscape: a near-monopoly at the frontier. TSMC holds ~62% of global foundry revenue and ~90% share at leading-edge nodes (sub-7nm). Samsung Foundry (~12%) is the only nominal competitor at the frontier but suffers chronic yield gaps. Intel Foundry Services remains sub-scale. GlobalFoundries, UMC, and SMIC compete only at mature nodes. The market is functionally a duopoly in name and a monopoly in practice at the technology frontier where margins and growth concentrate.

The value chain — EDA tools → fabless design → foundry → OSAT — funnels almost all advanced logic manufacturing through TSMC, making it the irreplaceable nexus between chip designers (Apple, NVIDIA, AMD, Broadcom) and end markets.

The market is a powerful, durable tailwind — AI capex is structural, not cyclical, and every dollar of it flows disproportionately through TSMC's fabs.

TSMC — Moat / Competitive Advantages

TSMC possesses what may be the widest moat in the semiconductor industry — a self-reinforcing flywheel of scale, process power, and capital intensity that has eliminated all but one credible competitor at the leading edge.

The core moat is process power compounded by scale. TSMC's manufacturing expertise — yield engineering, defect density control, and design-technology co-optimization — is the product of 30+ years and trillions of dollars in cumulative R&D and capex. This cannot be replicated by writing a check. Samsung has spent aggressively and still cannot match TSMC's yields at 3nm/2nm. Intel Foundry Services remains years behind. The knowledge is embedded in thousands of proprietary process recipes and a workforce of ~76,000 engineers.

Switching costs lock customers in. Chip designs are physically tied to a specific foundry's process design kit (PDK). Migrating a design to a rival foundry costs $50-100M+ and 12-18 months — with yield risk. For Apple, NVIDIA, AMD, and Qualcomm, switching is economically irrational when TSMC delivers superior yields and on-time execution.

Capital requirements create an almost impenetrable barrier. A single leading-edge fab costs $20B+; TSMC guided ~$38-42B in capex for 2025 alone. Only nation-state-level subsidies (CHIPS Act, EU Chips Act) make new entrants even theoretically viable, and even then they depend on TSMC to build or operate the fabs.

The moat is widening. Advanced packaging (CoWoS, SoIC/3DFabric) adds a second chokepoint — AI chips require these technologies, and TSMC dominates supply. The concentration of leading-edge revenue continues to intensify: 3nm and below represented a growing share of revenue in 2025, and the N2/A16 roadmap extends the lead through 2027+.

Financial Strength

TSMC's financial profile is elite — consistently among the best in the global semiconductor industry and across all capital-intensive businesses. The numbers speak clearly.

Returns on capital are exceptional and durable. FY2024 ROE reached ~30%, with ROIC in the mid-20s — roughly 2.5× any reasonable cost of capital estimate. These returns have been sustained across cycles: even in the soft 2019 market, ROE stayed above 20%. The gap to foundry peers (e.g., GlobalFoundries, SMIC) is enormous and widening as leading-edge economics favor concentration.

The balance sheet is a fortress. TSMC carried a net cash position of roughly NT$1 trillion at YE2024, even while running ~NT$1T/year capex. Debt-to-equity is negligible at ~0.3×. The escalation to $38–42B capex guided for 2025 is entirely self-funded from operating cash flow (~NT$1.94T in FY2024). This is investment for offense, not survival.

Earnings quality is clean. FCF conversion was ~81% of net income in FY2024 — lower than a typical asset-light business but remarkable for one investing at this intensity. Operating cash flow exceeds net income due to high depreciation, and receivables/inventory track revenue growth with no anomalous buildups. No goodwill, no acquisitions of note, no auditor changes (Deloitte & Touche, continuous). IFRS reporting is straightforward.

Customer concentration is the principal nuance. Apple at ~25% and the top-10 at ~70% of revenue is optically high — but switching costs are so extreme at leading-edge nodes that this concentration represents locked-in demand, not fragility.

I have sufficient training knowledge on TSMC's capital deployment. Let me write the analysis directly.

Runway for Reinvestment

TSMC possesses one of the longest reinvestment runways in global industrials, backed by returns on incremental capital that are rising, not declining. The company is deploying $38–42B in capex for 2025 alone — the largest single-year investment in semiconductor history — yet ROIC has expanded from ~25% to ~30%+ as leading-edge AI demand absorbs capacity before construction finishes. This is the hallmark of a business where demand growth outpaces even aggressive reinvestment.

The reinvestment opportunities are concrete and multi-year: N2/A16 node transitions, CoWoS advanced packaging expansion (the binding constraint on AI chip supply), and geographic diversification fabs in Arizona, Japan, and Germany. Each serves a structural demand vector — AI training/inference, sovereign chip security — not cyclical fill. The implied organic growth rate, combining ~30% ROIC with a ~60–70% reinvestment rate, points to mid-to-high teens sustained revenue growth.

Critically, TSMC does not waste capital on acquisitions or buybacks — nearly all retained earnings flow into high-return capacity. Historical return on incremental invested capital has exceeded 30% over the past five years, reflecting pricing power on leading-edge nodes and operating leverage as fabs mature.

TSMC's competitive position is not just strong — it is historically anomalous. No foundry peer comes close on the metrics that matter: technology leadership, margins, and customer lock-in.

The only meaningful competitor at leading edge (sub-7nm) is Samsung Foundry, which holds roughly 10% share versus TSMC's ~90%. Samsung has struggled with yields on 3nm GAA, losing customers (notably Qualcomm) back to TSMC. Intel Foundry Services remains pre-revenue for external customers at scale and is burning cash. GlobalFoundries, UMC, and SMIC compete only in mature/specialty nodes where TSMC also participates but pricing pressure is higher.

TSMC is gaining share, driven by AI accelerator demand concentrating at leading edge (N3/N2) where no one else can deliver at volume. The trend is structural, not cyclical — advanced packaging (CoWoS) adds a second moat layer that peers cannot replicate at scale.

The valuation gap reflects reality: TSMC trades at ~25x forward earnings versus GFS at ~20x and UMC at ~10x, but TSMC's margin structure and growth rate more than justify the premium. Samsung and Intel are spending aggressively to close the gap, but yield competitiveness remains years away — and TSMC continues to pull ahead.

Management & Shareholder Orientation

TSMC's governance is best-in-class for Asia and strong by any global standard. C.C. Wei, chairman and CEO since mid-2024, is a 30+ year company veteran who rose through engineering — a meritocratic succession from founder Morris Chang. Insider ownership is negligible in percentage terms given the ~$1T+ market cap, but this is structural for a company of this size, not a red flag. The Taiwan National Development Fund holds ~6.4%, providing a stable, aligned anchor. No share pledging has been reported.

The board includes genuinely independent directors and follows Western-standard governance practices uncommon for Taiwanese firms — audit, compensation, and nomination committees all function with independent oversight. Related-party transactions are immaterial and disclosed at market rates. No regulatory actions against the company or leadership.

Capital return is disciplined: ~30-35% payout ratio via steadily growing cash dividends (NT$17.5/share in FY2024, up from NT$8 in 2019), with the remainder reinvested in leading-edge capacity. No buybacks, which is standard in Taiwan. Buffett's 2022-23 purchase-and-sale was driven by geopolitical discomfort, not a business quality concern — a distinction worth noting.

Management Competence & Ethics

TSMC's management is among the most disciplined and transparent in global technology — a founder-built culture of operational excellence that has survived leadership transitions intact.

Capital allocation has been textbook value-creation. TSMC reinvests aggressively — capex scaled from ~$17B (2021) to ~$30B (2024) to ~$38-42B guided for 2025 — but only into leading-edge capacity where returns are structurally superior (gross margins expanded from ~52% to ~59% over that period). There are no value-destroying acquisitions on the record; TSMC's discipline is demonstrated by what it didn't buy. Dividends return roughly 25-30% of earnings, prioritizing reinvestment during the current capacity supercycle. No buyback programs of significance.

Execution against guidance is exceptional. Technology node transitions (N7→N5→N3→N2) have consistently delivered on or ahead of schedule. Revenue and margin guidance given on quarterly calls has been met or exceeded in the vast majority of quarters over the past decade. The Arizona fab (Fab 21) faced well-publicized early delays but management was forthright about the challenges and brought it largely back on track.

Transparency is a genuine strength. Quarterly calls provide specific revenue ranges, margin guidance, and capex figures. Management proactively discusses risks — CoWoS capacity constraints, geopolitical tensions, customer concentration — without being forced. Morris Chang's founder culture of candor persists under CEO C.C. Wei.

Clean record: The FY2025 20-F confirms no financial restatements, no auditor disagreements, and no error corrections. Audited continuously by Deloitte & Touche. The 2019 GlobalFoundries IP dispute was settled with cross-licensing; no material pending litigation threatens the business. No fraud allegations or whistleblower issues in the company's history.

TSMC — Valuation

TSMC is fairly valued at $2.22T — priced for excellence, not for perfection. The current multiple embeds ~18% earnings CAGR over five years, a demanding but achievable target given AI structural tailwinds and TSMC's monopoly position on leading-edge nodes.

Earnings power and embedded expectations. TTM net income is approximately $70B (implied by 31.9x trailing P/E on a $2.22T market cap). At a 10% discount rate and 22x terminal P/E, the market requires 2031 earnings of ~$163B — an 18% CAGR. TSMC exceeded this rate over the last three years (earnings nearly doubled from 2023 to 2025), but sustaining it requires AI chip demand to remain robust and margins to stay elevated despite costlier overseas fabs (Arizona, Kumamoto, Dresden).

Management credibility is high. TSMC has consistently beaten its own long-term revenue CAGR guidance of "close to 20%." Capital allocation is disciplined — ~26% payout ratio, minimal buybacks, and massive reinvestment ($40B+ annual CapEx) into capacity that is fully booked. The forward P/E of 19.6x reflects analyst consensus for ~$113B in forward earnings, implying 60%+ growth from TTM — aggressive but consistent with the current trajectory (77% TTM earnings growth).

Margins are at cyclical highs. Operating margin at 60.3% and gross margin at 64.2% are well above TSMC's 2019–2023 averages (~50% and ~53% respectively). Leading-edge AI pricing power supports elevated margins near-term, but geographic diversification structurally adds 3–5 percentage points of cost. Normalizing to 52% operating margin is prudent for terminal estimates.

Liquidation is irrelevant. Book value of ~$168B ($5.36T TWD) is a fraction of market cap. The value here is entirely in the going concern — irreplaceable manufacturing know-how, $100B+ in fabrication assets, and decade-long customer lock-in.

Probability-weighted expected value: ~$3.0T, suggesting ~35% upside over five years — a reasonable but unspectacular return for a best-in-class compounder. Not cheap enough for a wide margin of safety, but not expensive enough to avoid.

Long-term Valuation

TSMC's compounding engine is among the most powerful in global equities — high-return reinvestment into a widening moat, funded by a business that converts 60% of revenue to operating profit.

The reinvestment flywheel is self-reinforcing: each new process node (N2, A16) costs more to develop, raising barriers for followers while TSMC captures premium pricing from customers with no alternative. Capex of NT$1.28T in FY2025 earned 40% ROE on a rapidly growing equity base — incremental returns are not declining. AI/HPC now dominates the revenue mix, extending the runway well beyond smartphone cyclicality. Advanced packaging (CoWoS) adds a second compounding vector that is capacity-constrained through at least 2027.

What breaks it: The only credible threat to the 10-year thesis is geopolitical disruption to Taiwan operations — not competition. Intel and Samsung are 2–3 nodes behind with no visible path to parity. The thesis-breaking business signal would be a major customer (Apple, NVIDIA) successfully placing >20% of leading-edge volume elsewhere for two consecutive nodes.

At forward P/E ~20x on what is likely still an under-earning year, a 2–3x return over 10 years is plausible if the AI compute cycle sustains mid-teens revenue growth and margins hold above 55%. The risk-adjusted return is compelling — the geopolitical discount is real but the probability of permanent impairment remains low given ongoing geographic diversification (Arizona, Japan, Germany).

Risk Assessment — Taiwan Semiconductor Manufacturing Company Limited

TSMC's risk profile is dominated by a single tail event — a Taiwan Strait military conflict — while the business itself faces near-zero competitive or financial fragility.

Geopolitical concentration is the existential risk. Over 80% of TSMC's leading-edge capacity sits in Taiwan. A Chinese blockade or invasion would sever the world's most critical technology supply chain. TSMC's geographic diversification (Arizona, Kumamoto, Dresden) is underway but won't reach meaningful leading-edge scale outside Taiwan before ~2028-2030. The probability of a full military scenario in the next decade is low (estimated 5-10%) but the impact would be total — permanent capital impairment. TSMC's "silicon shield" role — where its destruction would devastate China's own tech supply — is itself a deterrent, but not a guarantee.

Business and competitive risks are negligible. No competitor is within two process generations of TSMC at leading edge. Intel Foundry and Samsung have spent tens of billions without closing the gap. Customer concentration (Apple ~25%, NVIDIA ~11% of FY2025 revenue) is mutual dependency — these customers have no viable alternative supplier. Technology obsolescence risk is near-zero on a 10-year horizon; silicon-based scaling continues via advanced packaging (CoWoS, SoIC) even if traditional node shrinks slow.

Financial risk is minimal. TSMC ended FY2025 with net cash, operating margins above 45%, and capex funded almost entirely from operating cash flow. Even a severe cyclical downturn would not threaten solvency.

Governance is clean. Post-founder transition has been seamless. No related-party concerns, no audit flags, no key-person dependency. Board includes independent directors with global semiconductor expertise.

Final Verdict: TSMC (TSM)

BUY — one of the best businesses in the world at a fair price, not a bargain.

TSMC is a generational compounder: ~90% leading-edge foundry share, 60%+ operating margins, 40% ROE, net cash balance sheet, and a self-funding capex machine reinvesting at rising incremental returns. The AI supercycle is structural, not cyclical — every major hyperscaler and chip designer depends on TSMC's 3nm/2nm nodes with no credible alternative. This is as close to a monopoly as exists in global technology.

The business passes every quality filter decisively. The question is price. At $2.22T / ~32x trailing earnings, the market already prices ~18% earnings CAGR. That's achievable — management guides "close to 20%" revenue growth with expanding margins — but it leaves limited margin of safety. The most-probable 2031 market cap of ~$3.1T implies ~7% annual capital appreciation plus ~1% yield, totaling ~8% annualized returns. Above-average for below-average business risk, but not a fat pitch.

Strongest counterargument (inversion): A Taiwan Strait military conflict is the only scenario that produces permanent capital loss. The probability is low but non-zero, and the consequence is binary — TSMC's fabs cannot be replicated quickly elsewhere despite geographic diversification into Arizona and Japan. This tail risk is why the stock doesn't deserve a STRONG_BUY even at a lower price.

Position sizing: Build in small tranches. The business deserves a core portfolio position, but the geopolitical tail risk and fair (not cheap) valuation argue against concentration. Accumulate on pullbacks toward 20-22x forward earnings.

For existing holders: HOLD. The thesis is intact, the moat is widening, and selling a business this good for valuation reasons alone is usually a mistake. Don't add aggressively at current levels — but don't trim either.

What to research further:

• Arizona fab yield rates and cost parity timeline — critical for de-risking the geopolitical thesis
• Customer concentration trends (Apple + Nvidia + AMD as % of revenue)
• Intel Foundry's 18A progress — the only plausible long-term competitive threat