Every leading-edge wafer is spoken for. The stock keeps asking whether the bet will yield.
TSMC just posted a record quarter — revenue +41% in dollars, EPS +69%, gross margin 66.4% — and raised 2026 growth to about 44%. The stock fell anyway, 12.5% from its Oct 5 high, because the market is pricing one question: does AI demand keep clearing a $60–64B-a-year capex bet, or is this peak yield? Six analyst lenses, three scenarios, four time horizons.
Gray line = TSM’s actual price from dated anchor prices (Dec 31 ’24 $197.49 → Apr ’25 low $134 → Jun 30 ’26 close $477.57 → Oct 5 ’26 close $485.80 high → $425.08 today); intermediate months are not drawn because dated closes were not available for them. Colored paths = synthesized scenario midpoints, probability-weighted (bear 30% · base 45% · bull 25%; bear-tilted because AI sentiment is crowded and capex is rising). Log price axis, mid-year marks. Clay circle = Wall Street 12-month consensus ≈ $584 (S&P Global, 21 analysts, range $440–$1,000; 14 Strong Buy, 6 Buy, 1 Hold, 0 Sell).
Re-weight the scenarios
Those probabilities are a judgment call — so make them yours. Drag to set how likely the bear and bull cases are (base takes the remainder); the blended target, the dotted line on the chart, the wafer yield map below and the prob-weighted row of the scenario cards all update live.
The yield map. One wafer, 100 dice, each die is one point of probability. Terracotta dice are bear-case outcomes, blue are base, olive are bull. Drag the sliders and the map re-cuts: the more the wafer yields olive, the higher the blended target above.
Six analyst lenses, six answers
The same fundamentals support wildly different conclusions depending on the framework. Each lens is a synthesized perspective with its own 12-month target (an explicit multiple on next-twelve-months EPS). Panel mean ≈ $471, range $273–$610.
The Toll Collector
Full-year 2026 revenue growth guided to ≈44% (up from “slightly above 40%” in July), Q4 guided at +58% YoY at the midpoint, and Street EPS compounding ≈35% into 2027. Management called demand “well above” supply and kept the AI-accelerator CAGR in the high-50s%. A franchise growing like this earns a premium multiple, not a de-rating.
The Cash Counter
Superb returns (ROIC ≈60%) and ≈$100B of net cash, but TTM free cash flow of $45.6B is only a ≈2.4% yield on the stock, a $60–64B capex budget is rising, and the dividend yield is ≈0.75%. Great business, fair price: the margin of safety is thin when the cash is all being re-invested.
The Capex Cliff
Record profit already failed to lift the stock: the Oct 14–16 AI selloff was set off by a post on CoreWeave’s debt, a Broadcom CEO casino analogy, and rate-hike worries. TSMC itself guides 2026–28 capex “significantly higher” than the prior three years and 2nm plus overseas fabs dilute gross margin 2–4 points. If customers’ AI spend digests, fixed costs bite.
The Only Foundry at the Frontier
2nm already reached 8% of wafer revenue (3% a quarter earlier), 3nm 33%, CoWoS demand “remained strong,” and gross margin of 66.4% printed inside guidance while the ramp dilutes it. Samsung is reported to be turning to TSMC for Nvidia’s next AI memory. The moat is process lead plus packaging plus trust, and no rival is close.
The Reset
Price sits between the 50-day ($434.50) and 200-day ($396.16) averages with RSI ≈42, 12.5% below the Oct 5 closing high. Forward P/E has compressed to ≈19.5x from ≈24.5x in January: earnings rose and the multiple fell. Mean reversion to ≈21x is the technician’s trade.
The Rate Strategist
Press coverage ties the selloff to rate-hike fears: Taiwan’s index had its worst week since April 2025, KOSPI fell nearly 9%, and a Bloomberg Asia chip index sat nearly 19% below its June peak. With tariff terms, an NT$ swing risk and Taiwan concentration on top, growth is real but the multiple is the swing factor.
Wall Street 12-month price targets
What the sell-side expects over the next year. Bars are sorted low to high; the dashed line is today’s $425.08 — seven of eight selected targets sit above it.
Eight of the most recent individual actions (Oct 15–19, 2026, TipRanks via StockAnalysis); the full consensus is $584.39 mean / $557.50 median across 21 analysts (S&P Global). Source conflict: S&P’s lowest target is $440, but TipRanks shows JLPascual downgrading to Sell with a $262 target on Oct 19 — we show both. Several desks raised targets after the print (Bernstein $430→$554, Needham $530→$560, Barclays $665→$670) even as the stock fell.
Where the road leads
Synthesized scenario midpoints (mid-year), each equal to next-twelve-months EPS × an exit multiple. Returns are versus today’s $425.08. These are illustrative frameworks, not predictions; five-year outcomes hinge on whether AI demand and TSMC’s pricing power hold.
1 Year
Mid-20272 Years
Mid-20283 Years
Mid-20295 Years
Mid-2031▸ Bull case — show the assumptions & math
▸ Base case — show the assumptions & math
▸ Bear case — show the assumptions & math
Revenue, capex, free cash flow & debt (NT$ trillions)
Where the money goes. The bull and bear theses both live in the gap between the capex and free-cash-flow bars.
Reported in NT$ to avoid currency distortion (S&P Global). 2023–25 are actual; lighter 2026E bars are estimates: revenue and FCF per consensus (NT$5.56T and NT$1.45T), capex at the $60–64B budget midpoint × NT$32 ≈ NT$1.98T. Total debt is gross (NT$0.82T at Sep 30, 2026) against NT$4.02T of cash and short-term investments, a net-cash business. Capex jumped from NT$0.96T to NT$1.27T in 2025 and is heading higher; revenue is growing faster, so FCF still compounds. The bear’s case is that the capex bar keeps climbing after revenue stops.
EPS path underpinning the targets ($ per ADR)
Each price target is an EPS estimate times an exit multiple. Here is the earnings ladder the scenarios are built on.
One ADR = 5 ordinary shares; converted at a constant NT$32 per US$ (Q3’26 EPS of NT$29.43 = US$4.60 per ADR implies 31.99), so growth is not a currency artifact. Gray = reported (2024 derived from net income ÷ 25.93B shares); olive = estimates. 2026E–2027E are Street (NT$110.13 and NT$148.61, S&P Global); starred 2028E–2031E are our base case. The base ≈ $47 of 2031 EPS at ≈18x ≈ the $874 five-year base target.
The business is still accelerating
Q3 FY26, year-over-year — read these against a stock 12.5% below its high.
Every line is above +40%, with profit growing faster than revenue. Olive = reported Q3 lines (TSMC release, Oct 15, 2026; operating-profit and FCF growth per Hargreaves Lansdown’s summary of the results); clay = Q4 guidance midpoint (+58% YoY, CNA) and free cash flow, which more than tripled to ≈$14.9B. FCF bar capped at +200% for display. Net margin is 51.1%. The disconnect between this chart and a falling price is the bull case.
Bull vs. Bear
The valuation argument compresses into one disagreement: is TSMC the toll road every AI dollar must cross, or a capex-hungry cyclical at peak yield?
▲ THE BULL CASE
- Record quarter, raised outlook. Revenue $46.72B (+41%), above the top of guidance; 2026 growth now ≈44% versus “slightly above 40%” in July.
- Q4 guide is a step up. $52.3–54.3B implies +58% YoY at the midpoint and roughly 14% sequential growth.
- Pricing power survives the ramp. Gross margin 66.4% (inside the 65–67% range) while 2nm went from 3% to 8% of wafer revenue.
- Demand exceeds supply. CoWoS packaging stays tight; management says demand remains “well above” capacity and sees “no sign of a slowdown.”
- Fortress balance sheet. ≈$100B net cash, debt/EBITDA 0.23x, ROIC ≈60%, and a dividend that grew 32% last year.
- The multiple already reset. Forward P/E ≈19.5x versus ≈24.5x in January, in line with the S&P 500 at ≈20.3x and with Nvidia at ≈19.9x, for a company growing EPS 35%+.
▼ THE BEAR CASE
- Capex is a one-way ratchet. $60–64B in 2026 (from $52–56B in January); 2026–28 guided “significantly higher” than the prior three years. TTM FCF yield is only ≈2.4%.
- Good news no longer lifts the stock. TSM fell 1.8% on the day of a beat-and-raise and 11.5% in seven sessions as the AI trade unwound on debt and rate worries.
- Margin dilution is guided. 2nm dilutes gross margin 2–4 points and overseas fabs another 2–4; Q3 margin slipped to 66.4% from 67.7%.
- Customer concentration. HPC is 68% of revenue; a hyperscaler or neocloud pause (the CoreWeave worry) hits the biggest line directly.
- Taiwan concentration and policy. Most leading-edge capacity is on one island, and chip-tariff terms with Washington keep shifting.
- The multiple is not cheap on trailing numbers. 27.8x trailing, 8.6x book and 12.3x sales; a JLPascual Sell at $262 shows where the bear’s math lands.
Risk map — likelihood × impact
Where each risk sits, not just how big it is. The hot corner — likely and high-impact — is AI sentiment, which has already moved the stock; the worst-case tail (a Taiwan Strait shock) sits in the Tail row by design.
- NT$ swings
- Margin dilution
- Capex creep
- AI de-rating
- Rival 2nm share
- Arizona costs
- AI capex digestion
- US chip tariffs
- Taiwan Strait shock
AI-sentiment de-rating
It just happened: a CoreWeave-debt post and a Broadcom casino comment cut TSM about 11% in seven sessions, and the forward multiple is already down from ≈24.5x to ≈19.5x. Another leg lower in AI sentiment moves the stock more than earnings do.
AI capex digestion
Hyperscalers and neoclouds pause orders after a record build. With HPC at 68% of revenue and capex at $60–64B, utilization slips while depreciation keeps rising.
US chip tariffs
Washington has threatened steep duties on foreign-made chips; a US–Taiwan deal reportedly lowers rates in exchange for US investment. A reversal re-prices cost and customer demand.
Margin dilution
Management guides 2nm to dilute gross margin 2–4 points and overseas fabs 2–4 more; Q3 margin already slipped to 66.4% from 67.7%.
Capex creep
The 2026 budget went from $52–56B to $60–64B in six months, and 2026–28 is guided higher still. Free cash flow yield stays thin.
Taiwan Strait shock
A blockade, conflict or severe quake at the Hsinchu and Tainan clusters: low odds, but most leading-edge capacity sits on one island.
Rival 2nm share
Samsung and Intel foundry win a marquee 2nm or A16-class design, capping pricing power at the margin.
Arizona costs
A $265B US build-out (3nm Fab 2 now complete) brings higher costs and lower yields before it reaches scale.
NT$ swings
A stronger Taiwan dollar erodes margin on NT$ costs against US$ revenue; EPS in NT$ and US$ can diverge.
The jargon, decoded
Hover the dotted terms in the metrics, or scan the desk’s working definitions here.
- ADR unit
- The NYSE-listed certificate worth 5 ordinary Taiwan shares. All per-share figures here are per ADR, converted at NT$32 per US$.
- 2nm / 3nm (N2 / N3)
- Chip-making process generations; smaller is faster and more efficient and costs more to build. 2nm was 8% of wafer revenue in Q3, 3nm 33%.
- CoWoS
- TSMC’s advanced packaging that links AI processors to memory. It is the current bottleneck for AI accelerators.
- Gross margin
- Revenue less cost of making the wafers, as a share of revenue. Ramping 2nm and overseas fabs dilutes it by a few points.
- Capex
- Money spent on fabs and tools. $60–64B in 2026; it is what separates operating cash flow from free cash flow.
- Free cash flow yield
- Free cash flow ÷ market value. About 2.4% here: the cash is large, but the stock price is larger and capex is rising.
- Exit multiple
- The forward P/E assumed at the end of a forecast. Multiply it by next-twelve-months EPS to get the scenario price.
- Prob-weighted
- Each scenario’s price × its probability, summed into one expected value across bear, base and bull. Use the sliders to change the weights.