TSM — Research Report (2026-07-14)

TL;DR

• What it is: The world's largest contract chip factory ("pure-play foundry") — it manufactures chips designed by Apple, Nvidia, AMD and others, but sells nothing under its own brand.
• Overall take: A genuine monopoly on leading-edge chip-making, firing on AI demand, priced richly but not absurdly — the risk is a map coordinate, not the business.
• Revenue $122.4B in 2025, +35.9% YoY (growth is accelerating, not fading).
• Net margin 45.1%, gross margin 59.9% (rare pricing power for a factory).
• ~64% of the global foundry market (structural monopoly at the leading edge).
• Trailing P/E ~36×, forward ~27× (priced for growth, cheaper on next year's earnings).
• Strongest bull point: It is the sole viable maker of the most advanced AI chips — Nvidia and Apple have nowhere else to go.
• Strongest bear point: ~90%+ of leading-edge capacity sits in Taiwan; a China–Taiwan conflict could reprice the stock regardless of earnings.
• Watch next: Q2 2026 results on July 16, 2026 — specifically whether gross margin holds ≥65% while overseas (Arizona/Japan) fabs dilute it.

1. Business overview

So what: TSMC is the neutral factory the entire chip industry rents — it builds everyone's chips and competes with none of its customers, which is exactly why it wins.

TSMC is a pure-play foundry — "foundry" means a factory that manufactures chips other companies designed; "pure-play" means that's all it does (unlike Intel or Samsung, which also design and sell their own chips). Its customers — Nvidia, Apple, AMD, Qualcomm, Broadcom — hand over blueprints; TSMC turns silicon wafers into finished chips.

That neutrality is the whole moat: a fabless designer ("fabless" = designs chips, owns no factory) can trust TSMC not to steal its designs or favor a rival, which it could never fully trust from Samsung or Intel.

Revenue is concentrated in advanced nodes. A "node" (e.g. N3 = 3-nanometer, N5 = 5nm) measures how small the transistors are — smaller means faster and more power-efficient, and only TSMC currently makes the very smallest at scale. In 2025, chips at 7nm-and-below were 74% of wafer revenue (up from 69% in 2024) — this is the high-margin, hard-to-copy end. Q4'25 platform data

Source: Futurum / TSMC Q4'25. AI accelerator revenue alone was "high-teens %" of total and is the growth engine.

Geography: designed-in-Taiwan production, sold globally. Overseas fabs are ramping in Arizona (US), Japan, and Germany — partly customer demand, partly geopolitical insurance.

2. Financials (3–5 year)

So what: Revenue more than doubled in four years while margins rose — this is a capital-heavy factory that behaves like a software company on profitability.

All figures USD, from TSMC's official releases (FY25 6-K, FY23/24 6-Ks):

• 4-year revenue CAGR ≈ 21% (CAGR = the smoothed annual growth rate). 2023 was the one down year — a normal chip-industry inventory downcycle, not a broken business.
• The margin slope is the tell: gross margin climbed from ~51% to 59.9% even as the company spent record amounts on new factories. That is pricing power in motion.

Cash flow reality (profit is opinion, cash is fact)

Sources: 2021–24 FCF, FY24 capex $29.76B, FY25 capex NT$1,272B, FY25 FCF ~$31.7B.

Key point: FCF (cash left after building factories) is positive and rising despite ~$40B/year of capex. Cash conversion (FCF/net income) ≈ 57% — the rest is eaten by fab construction. This is the crux: TSMC must spend enormously to stay ahead, and it still throws off ~$30B of surplus cash.

Balance sheet

Net cash ≈ $71.6B ($105.8B cash vs $34.2B debt) as of end-2025. Source. A capital-heavy business with net cash (more cash than debt) has essentially zero financial fragility — it could fund a downturn from its own pocket.

3. Metrics table — with the "why this matters here"

So what: every number below is high-quality; the only debatable one is the price you pay for it.

ROE and share figures: TSMC ownership/share data; capex guide: FY25 release.

4. Unit economics — the one number the business runs on

So what: TSMC's economics run on wafer price × node mix — richer nodes command multiples of the price, and TSMC owns the richest ones outright.

The foundry's unit is a wafer (a disc of silicon that gets cut into hundreds of chips). The number that runs the business is average wafer price, driven by node:

• A leading-edge N3 (3nm) wafer sells for a reported ~$18,000–$20,000+, vs ~$3,000 for a mature (28nm) wafer — roughly 6× the price for the newest node.
• So every point of mix-shift toward N3/N5 lifts revenue and margin without a single extra wafer. In 2025, N3 was 24% of wafer revenue and N5 ~35% — the mix keeps marching to the high-price end.
• TSMC also raises prices on leading-edge and advanced-packaging (CoWoS, the technology that stitches AI chips + memory together) because there is no alternative supplier — 2026 leading-edge price hikes are widely reported.

The lazy way to read the whole business: are the newest, priciest nodes a growing share of wafers? In 2025, yes — decisively. Node mix source

5. Ownership & management

So what: no controlling family or founder — it's a widely-held institutional stock with a symbolic government anchor and a low-drama, engineer-led board.

Source: TSMC ownership.

Management: Dr. C.C. Wei is Chairman & CEO (took the combined role after Mark Liu's 2024 retirement). A career TSMC engineer, not a dealmaker — the culture is operational excellence, not empire-building. No dominant insider ownership means no key-man risk but also no founder-with-skin-in-the-game signal; incentives run through conventional pay and options. Dividend: a steady, rising quarterly cash dividend (~TWD 4.5–6/share per quarter in 2025) — a shareholder-friendly capital-return habit, though the yield is modest because most cash goes back into fabs. Dividend data

Insider signal: none material or findable — consistent with a widely-held, professionally-managed large cap. (Not a red flag; simply no strong buy/sell signal to read.)

6. Market & value-chain position

So what: TSMC is the bottleneck of the AI economy — everyone above and below it depends on it, and almost no one has pricing power over it.

• Position: definitional bottleneck, not a commodity. It sits between chip designers (Nvidia, Apple) and chip equipment makers (ASML, Applied Materials). Designers have nowhere else to make cutting-edge chips at volume; that's why TSMC can raise prices.
• Who has pricing power over TSMC? Only one company: ASML, the Dutch monopoly on EUV lithography machines (the ~$200M+ machines that print the smallest transistors — "EUV" = extreme ultraviolet light). TSMC is ASML's biggest customer, but ASML is a true single-source supplier, so pricing power there runs upstream. Everyone else — customers, packaging, materials — is a price-taker to TSMC.
• Competitors:
  • Samsung Foundry — the only other company making leading-edge logic, but with lower yields and trust issues (it competes with its own foundry customers). Distant #2.
  • Intel Foundry — trying to become a foundry with US-government and Nvidia/customer backing, but years behind on volume yield.
  • Rapidus (Japan) — a state-backed 2nm startup; a 2027+ story, not a 2026 threat.
• Share: TSMC holds ~60–64% of the entire foundry market and a far higher share of the leading edge (~90%+ of the most advanced nodes). Share source

This is the §0 lens from the research playbook: TSMC is the monopoly bottleneck of a macro-underwritten trend (AI). You rarely find the bottleneck this profitable and this exposed to one risk at the same time.

7. Valuation

So what: On next year's earnings TSMC trades roughly in line with big semis and below the AI-hype names — expensive vs its own history, but not "priced for perfection" given the growth.

• Price ~$434 (July 10, 2026), market cap ~$1.99T. Source

Sources: trailing/forward P/E, forward P/E vs industry. Sector ranges from the same screens; EV/EBITDA derived (EV ≈ $1.92T after net cash / EBITDA proxy ≈ $87B). Which multiple to use: P/E — TSMC is highly profitable and net-cash, so earnings-based multiples are the right yardstick; EV/EBITDA is shown for the capital-heavy cross-check.

What a reasonable buyer might pay: At forward P/E ~27× on a business growing revenue ~30% with 45% net margins and a monopoly moat, the PEG (price/earnings-to-growth) is under 1 — cheap if the growth holds. A disciplined buyer would happily pay the current multiple for the earnings and treat the geopolitical discount as the reason it isn't 35×. The number that makes this stock cheap or expensive is not the P/E — it's the probability you assign to a Taiwan disruption.

Bull / base / bear (each attributed to a named institution)

Catalysts (bull): Q2'26 print (Jul 16), 2nm (N2) volume ramp in H2 2026, sovereign-AI orders, CoWoS packaging sold out. What would prove the bull thesis wrong: (1) gross margin breaking below ~55% as overseas fabs dilute; (2) an AI-capex air-pocket where Nvidia/hyperscalers pause orders; (3) any material China–Taiwan escalation — the one risk no order book can offset.

📚 What this company teaches

1. Find the bottleneck, not the brand. The AI trade's household name is Nvidia, but every AI chip — Nvidia's included — is made by TSMC. The most defensible profits often sit one layer below the famous end-product, at the choke point the whole value chain must pass through. (This is exactly the §0 "drill down the value chain" lesson: buy the picks-and-shovels monopoly, not just the miner.)

2. A factory can have software margins — if it's a monopoly on the hard part. Capital-heavy businesses usually earn thin margins because anyone can build a factory. TSMC earns 60% gross margins because no one else can build its factory (the yield know-how takes a decade). Capital intensity is only a curse when it isn't a barrier.

3. Cash conversion is the tax on a growth machine. TSMC's net income and its free cash flow diverge sharply (~57% conversion) because staying ahead costs ~$40B/year in new fabs. When you own a capital-heavy compounder, watch FCF, not just earnings — the day capex stops being reinvestable growth and becomes a maintenance treadmill is the day the story changes.

Self-test question: TSMC's forward P/E (~27×) is lower than its trailing P/E (~36×) and lower than the semiconductor industry median (~35×). What single assumption is the market making that, if wrong, would make the stock look expensive rather than cheap? (Hint: it's baked into the word "forward.")

📖 Glossary

• Foundry — a factory that manufactures chips designed by other companies; it doesn't design chips itself.
• Pure-play foundry — a foundry that only manufactures for others and never sells its own branded chips (TSMC's model; the source of customer trust).
• Fabless — a chip company that designs chips but owns no factory (Nvidia, Apple, AMD, Qualcomm).
• Node (e.g. N3, N5, 7nm) — a measure of transistor size; smaller = faster and more power-efficient. Only the newest nodes command premium prices.
• Wafer — a disc of silicon that a foundry processes and then cuts into hundreds of individual chips; the foundry's basic unit of sale.
• Advanced node / leading edge — the smallest, newest, hardest-to-make nodes (≤7nm), where TSMC has near-total share.
• EUV (extreme ultraviolet lithography) — the technology (and ~$200M machines from ASML) used to print the smallest transistors; ASML is the sole supplier.
• CoWoS — TSMC's advanced packaging technology that stitches an AI chip together with memory; currently sold out and a pricing lever.
• HPC (high-performance computing) — TSMC's largest segment: AI accelerators, server CPUs/GPUs — the AI-boom bucket.
• P/E (price-to-earnings) — share price ÷ earnings per share; how many dollars you pay per $1 of annual profit. "Forward" P/E uses next year's expected earnings.
• Forward P/E — P/E calculated on projected future earnings; lower than trailing P/E when earnings are expected to grow.
• EV/EBITDA — enterprise value (market cap + debt − cash) ÷ EBITDA (earnings before interest, tax, depreciation, amortization); a debt-neutral way to price the whole business.
• FCF (free cash flow) — operating cash flow minus capital spending; the real cash left over after keeping the business running and growing.
• Cash conversion (FCF/net income) — what fraction of accounting profit turns into actual spendable cash; low here because of heavy fab spending.
• Capex (capital expenditure) — money spent building/equipping factories; TSMC's is ~$40–56B/year.
• Net cash — cash on hand minus total debt; positive means the company owes less than it holds (financial safety).
• ROE (return on equity) — annual profit ÷ shareholder equity; how hard the owners' money works.
• CAGR (compound annual growth rate) — the smoothed average yearly growth rate over a multi-year span.
• PEG ratio — P/E divided by the growth rate; under 1 suggests growth is cheap relative to the price.

Sources are linked inline. Primary financials from TSMC's own SEC 6-K releases and investor materials; valuation and analyst targets from cited data providers and named banks (Goldman Sachs, Morgan Stanley, JPMorgan). FY25 operating cash flow and EV/EBITDA are labeled estimates derived from reported OCF/capex/net-cash figures. Not financial advice — a personal research note.