brain/
conceptstock-market

TSMC capacity shortfall and 4-year pricing power

Notes

TSMC capacity shortfall and 4-year pricing power

One-line summary: TSMC CEO C.C. Wei publicly acknowledged capacity is "about three times short" of customer plans (Nov 20 2025), and customers face price increases for four consecutive years starting 2026 — bullish for TSM margins, but the same conditions are the forcing function driving customers toward alternative foundries (see us-fab-capacity-bottleneck).

The insight

Two trades exist simultaneously and both can work:

  1. TSM long, on supply discipline + pricing power — sustained margin expansion through 2030 driven by AI-demand-led pricing.
  2. Alternative-foundry long (Intel) — the same customers who pay TSMC's higher prices are also the customers signing Intel anchor deals to diversify away.

These are not contradictory. The bullish read on both is that AI silicon demand is so far above supply that multiple winners can absorb capacity at premium pricing for years.

The chain

TSMC 3x-short capacity → 4-year price-hike cadence → marginal buyers shift to Intel 18A/14A → anchor customers (AWS, MSFT, Apple, Terafab) stack → INTC re-rate, plus picks-and-shovels (ASML/AMAT/KLA/LRCX) capture both sides.

Canonical: tsmc-saturation-to-intel-anchor-stack.

Evidence

Q2 FY2026 call (2026-07-17 ingest) — the shortfall restated, still unquantified

  • The gap is confirmed but TSMC refuses to size it. Asked directly by Robert Sanders (Deutsche Bank) whether unconstrained ≤3nm demand runs "30%-50% above your ability to supply... or something even larger," cc-wei in 2026-07-16-earnings-tsm-q2-fy2026: "No, we don't have a number to share. Let me say that the gap is very big." The refusal is itself informative — the analyst's 30–50% framing was offered and not corrected downward.
  • Every quantitative guide moved up, and none was quantified. cc-wei in 2026-07-16-earnings-tsm-q2-fy2026: on the five-year AI CAGR (prior: mid-to-high 50%), "let me give you not a number, but it's stronger and stronger... Stronger than what we said before." On the Symposium's 2nm-family ~70% capacity CAGR: "Okay, now is bigger. That's all I say." On out-year revenue: "Next few years is going to be a very good business for TSMC. That's all I can say."
  • Duration extended to 2029–2030 — cc-wei in 2026-07-16-earnings-tsm-q2-fy2026: "I believe from this day on all the way to probably 2029, 2030, the demand is very strong. Whether in between there's a dip or not, I'm not very sure. The trend is so robust that I believe we are witnessing a kind of a new industry."
  • Pricing power is real but self-limited — this bounds the thesis. cc-wei in 2026-07-16-earnings-tsm-q2-fy2026: "We are a partner... our customer got to be successful. I don't want to squeeze them out from the market... We don't suddenly increase our price by, which I like to have, a 4x or 5x." And the tell that TSMC believes it is under-pricing: "Yes, I'm really jealous about memory companies, 86% gross margin. 86%? About 68%, I would be happy about that." A monopolist targeting ~68% GM while a supplier upstream earns 86% is choosing customer retention over extraction — which caps the pricing-power upside this concept can claim.
  • The supply response is more expensive than the capex line implies — capex raised to $60–64B, but cc-wei names tool-price inflation as one of the two drivers ("we buy the tools with inflation price"). See semicap-tool-price-inflation: a rising capex number is partly price, not capacity, which lengthens the shortfall per dollar spent.
  • The binding constraint has moved to the back end — cc-wei in 2026-07-16-earnings-tsm-q2-fy2026: "our packaging capacity is so tight that now it's limiting my customers' growth." The shortfall is no longer primarily a wafer story. See cowos-packaging-capacity-crunch.
  • A14 confirms the structural lead time — cc-wei in 2026-07-16-earnings-tsm-q2-fy2026: "The lead time to develop a new technology such as A14, building the capacity, and then ramping it up now takes five to seven years. There are no shortcuts." Pre-production 2027, volume 2028.

Implications for trade structure

  • Supply discipline is years-long: pricing power doesn't compress in 12 months even if AI demand normalizes.
  • Customer dependency is fraying at the top: Apple, AWS, Microsoft, and Musk are all publicly sourcing alternatives. This isn't dispersed grumbling — it's the largest customers acting at the same time.
  • TSM and INTC can both work if you size for the different time horizons (TSM: now-2028 margin compounding; INTC: re-rate event over 12–24 months as deals convert).

Contradictions / tensions

  • The same fact pattern (rising prices + capped supply) is bullish TSM and bearish TSM (loses long-term customer concentration). Sequencing matters more than direction.
  • Wei's "T-shirt" anecdote could be read as confidence (we have all the demand we want) or as warning (don't expect us to deliver) — the truth is probably both, and it depends on which customer's perspective you take.

What would weaken this thesis

  • TSMC announces dramatically accelerated US capacity (e.g., 3nm Arizona pulled forward to 2026 from 2H27)
  • Pricing strategy reverses (unlikely, but a competitive response if Intel/Samsung win meaningful share)
  • AI silicon demand normalizes faster than expected (unlikely on current evidence)

Open questions

Valuation snapshot

Last refreshed 2026-07-20 (pre-open; marks are the Friday 2026-07-17 close, markets closed over the weekend). Price fills tagged twelvedata. Mkt cap / Fwd P/E are not in the Twelve Data free tier — sourced from stockanalysis.com on 2026-07-17 (tagged stockanalysis). Note the prior snapshot's "~$2.26T" market cap does not reconcile with the $1.84T sourced today; the older figure was undated and is discarded rather than carried.

TickerPrice52w rangeMkt capFwd P/EDay / vs 52w hiWhat's priced in (one line)
TSM$398.37$223.70–$479.00$1.84T stockanalysis18.58× stockanalysis−2.77% day; −16.8% from hiThe tape sold the print. TSMC raised the FY26 AI-growth guide to "slightly above 40%," raised capex to $60–64B, and announced an additional $100B for Arizona — and the stock fell 2.3% into a broad semis risk-off (SOXX −4.5%). Priced: pricing power, Arizona year-one profitability, the Vera Rubin lock-in. Newly un-priced, and it cuts the other way: cc-wei in 2026-07-16-earnings-tsm-q2-fy2026 disclosed that TSMC is choosing not to extract — "I'm really jealous about memory companies, 86% gross margin... About 68%, I would be happy about that" — which caps the margin the pricing-power thesis can claim

Forward-looking outcomes (12-month)

Bull case — the duration extension gets priced: cc-wei in 2026-07-16-earnings-tsm-q2-fy2026 extended the demand horizon to "from this day on all the way to probably 2029, 2030, the demand is very strong... we are witnessing a kind of a new industry," and moved every guide up without quantifying any of them — the five-year AI CAGR is "stronger and stronger," the 2nm-family capacity CAGR is "bigger," and the ≤3nm supply gap is "very big" (Wei declined to correct an analyst's 30–50% framing downward). If the market re-rates from a cyclical foundry to a five-year structurally-short one, the multiple does the work. Implied price: +25–40% from current.

Base case — earnings compound, the multiple doesn't: the guides are up and the book is booked (N3 through 2028, N2 sold out, CoWoS through 2027 — 2026-05-20-autoresearch-tsmc-capacity-pricing-power-n2-cowos-may-2026), but three disclosed drags cap the margin story. wendell-huang in 2026-07-16-earnings-tsm-q2-fy2026 puts overseas-fab gross-margin dilution at 2–3% early, widening to 3–4% "in the next several years"; part of the raised capex buys tool-price inflation rather than wafers (semicap-tool-price-inflation); and Wei's own ~68% GM target is a choice to under-price. Implied price: +10–20% from current.

Bear case — the self-imposed ceiling is the bear case, and it is new — the classic bear leg was share loss (Samsung SF2P, Intel 18A, Apple dual-sourcing — 2026-05-18-autoresearch-samsung-foundry-as-third-alternative-may-2026, 2026-05-11-autoresearch-intel-18a-yield-vs-tsmc-samsung-2026). Today's call adds a self-inflicted one: cc-wei in 2026-07-16-earnings-tsm-q2-fy2026 — "We are a partner... our customer got to be successful. I don't want to squeeze them out from the market... We don't suddenly increase our price by, which I like to have, a 4x or 5x." A monopolist that publicly commits to under-extracting is telling you the ceiling on its own margin, and it does so while conceding it earns 68% against a supplier upstream at 86%. Layer on the disclosed overseas dilution and tool inflation and the pricing-power thesis frays from the inside rather than from competition. Implied price: −10–20% from current.

Currently undervalued vs base case? Marginal, and the read tightened today rather than flipping. At 18.58× forward (stockanalysis, 2026-07-17) TSM is not expensive for a business guiding FY26 AI growth "slightly above 40%" — and the stock fell 2.3% on the raise, which is the tape refusing to pay for the duration extension. But the thing that would justify a genuine re-rate — margin expansion on monopoly pricing — was explicitly disclaimed by the CEO on this call, and the overseas-dilution and tool-inflation drags are now quantified against it. The honest read: cheap for the growth, correctly priced for the margin TSMC has chosen to accept. Upside remains multiple re-rating on Taiwan-risk resolution, not on extraction.

Catalyst path:

  • Q3 2026 earnings (October) — the first read on whether the $60–64B capex and the "slightly above 40%" FY26 guide hold; watch the GM line against Wei's stated ~68% comfort level.
  • Any dated schedule for the $100B Arizona commitment — Wei: "If you ask me to give you a firm schedule, no, we don't have it today." An unscheduled $100B is a 2030s supply event on his own five-to-seven-year node lead time; a date would make it real. See us-fab-capacity-bottleneck.
  • Apple September product cycle — any Intel/Samsung sourcing mention tests the share-loss leg.

Related

Referenced by
Mechanisms
Agentic AI → CPU resurgence in AI datacenters → additive silicon demand, ISA-agnostic → TSM captures it whoever wins; server-CPU scarcity is the corroborating tellSamsung SF2P yields hit 70% + Taylor 90% ready + Apple visit → genuine third foundry leg → order flow splits, not concentratesTaiwan energy blockade → island browns out in <3 weeks → fabs power down → 90-day restart → global shock, no shot firedTSMC packaging scarcity caps customer growth → strands TSMC's own (larger) front-end wafer revenue → incumbent welcomes a second-source packager → Intel EMIB-T demand is validated by the monopolist itselfTSMC saturation → 4-year price hikes → marginal buyers shift → Intel anchor stack → re-rate
Concepts
Agentic AI → CPU resurgence → incremental silicon demandCopper supercycle: AI data centers and energy buildoutCoWoS packaging capacity crunchHBM supply bottleneckMature-node shortage is confined to AI-adjacent segments (PMIC + sensors)Semicap tool-price inflation: capex dollars ≠ capacityUS fab capacity bottleneck → Intel as the alternative foundry
Questions
Hypothesis: Arm Holdings as Vera CPU architecture licensor → datacenter royalty upside not priced inHypothesis: Section 232 Phase 2 tariffs → US chip fab structural premium pricingHypothesis: TSMC CoWoS overflow → Amkor Technology as uncaptured OSAT beneficiaryHypothesis: TSMC Kumamoto → Emerging Physical AI Hub for Automotive/RoboticsIs the $830B CSP CapEx cycle a one-year spike or a sustained multi-year structural level?
Entities
Gavin BakerGavin UbertiMicron TechnologyPat GelsingerRob WachenTSMC (Taiwan Semiconductor Manufacturing Co.)Wendell Huang