The consensus way to own TSMC is as a volume story: more AI accelerators, more wafers, more revenue. Its own second-quarter numbers say something different. Wafer shipments rose 16.6% year on year in the three months to 30 June 2026, to 4,336,000 twelve-inch equivalents, while dollar revenue rose 33.7% to US$40.20 billion, according to the quarterly presentation TSMC filed with the SEC on 16 July 2026. Divide one by the other and blended revenue per wafer went from roughly $8,087 in Q2 2025 to roughly $9,271 in Q2 2026 — a rise of about 14.6%. Close to half of TSMC’s growth last quarter came from price and mix rather than output, and that is before the 5% to 10% list increase the foundry has negotiated for January 2027. Taiwan Semiconductor Manufacturing Company (NYSE: TSM) closed at $414.00 on 1 September 2026. Our TSMC stock prediction runs a $610 bull case against a $265 bear case, and both are built on that one ratio.
The number nobody is quoting
Having tracked foundry disclosures through three capex cycles, the ratio I check first is not revenue growth — it is revenue growth divided by shipment growth. It tells you whether a manufacturer is selling more things or charging more for the same things. For most of the past decade TSMC’s answer was “more things.” In the June quarter the answer was “both, in almost equal measure.”
That matters because price-led growth and volume-led growth carry different risk. Volume growth is sticky: a customer that has taped out a design on N3 cannot move it without an 18-month redesign. Price growth is not sticky at all. It is the first concession a supplier grants and the last one a customer forgets. The nearest parallel is not another chipmaker but a landlord in a housing market with no new supply: rents ratchet upward for as long as the queue at the door is longer than the number of doors, and the profit-and-loss statement looks unbreakable right up to the month the queue shortens.
So the bull and bear cases on TSM are not really an argument about AI demand. Both sides agree demand is enormous. They are an argument about how much of the 2026 income statement is rent, and what multiple you pay for earnings whose composition has quietly changed.
Key facts
- Q2 2026 revenue was US$40.20 billion, up 33.7% year on year, with a gross margin of 67.7% and an operating margin of 60.3% — TSMC earnings release, 16 July 2026
- Wafer shipments were 4,336,000 twelve-inch equivalents, up 16.6% year on year, implying blended revenue per wafer of about $9,271 versus about $8,087 a year earlier — TSMC 2Q26 presentation, 16 July 2026 (FinanceFeeds calculation)
- Advanced technologies — 7nm and below — were 77% of wafer revenue, with 2nm at 3%, 3nm at 30% and 5nm at 33% — TSMC earnings release, July 2026
- TSMC’s ten largest customers were 78% of net revenue in 2025, up from 76% in 2024 and 70% in 2023 — TSMC Form 20-F, filed 16 April 2026
- 2026 capital expenditure was set at US$52–56 billion in the annual report, and the board approved a further US$29.44 billion of capital appropriations on 11 August 2026 — Form 20-F and TSMC board resolutions, 11 August 2026
- July 2026 revenue was NT$467.58 billion, up 44.7% year on year, taking January–July to NT$2,872.06 billion, up 37.0% — TSMC July 2026 revenue report
- Nineteen analysts carry an average target of $554.45, with a high of $700 and a low of $440 — stockanalysis.com, 2 September 2026
What the second quarter actually showed
TSMC guided Q2 2026 to revenue of US$39.0–40.2 billion and a gross margin of 65.5% to 67.5%. It landed at the very top of the revenue range and above the top of the margin range, at 67.7%. Net income was NT$706.56 billion, up 77.4% year on year — more than double the rate of revenue growth, which is what operating leverage looks like when a fab base built for one demand curve gets used by a steeper one. Return on equity annualised at 45.9%.
The node mix explains the margin. Two-nanometer went into volume production in the fourth quarter of 2025, according to TSMC’s own process documentation, and reached 3% of wafer revenue within two quarters. Three-nanometer is 30%, five-nanometer 33%. Together, 7nm-and-below is 77% of wafer revenue — a record. Mature nodes, the part of the business most exposed to Chinese competition, are now a rounding error at the margin line: 28nm is 6%, 16/20nm 6%, 65nm 4%.
The pricing layer sits on top. TSMC is raising production prices by as much as 10% from the start of 2027, with base increases of 5% to 10% and a further 10% to 15% premium available on high-performance computing capacity booked above a customer’s original forecast — a second tier of pricing for late AI orders, as FinanceFeeds reported in July. Negotiations ran from June to July 2026. TSMC declined to confirm specific figures, saying only that its pricing strategy is “strategic, not opportunistic.”
Two balance-sheet lines deserve more attention than they got. Inventory turnover days rose to 87 in Q2 2026 from 76 a year earlier, and accounts-receivable days to 29 from 23. Neither is alarming on its own. But working capital lengthening by 17 combined days while revenue accelerates is the sort of thing that reads as demand pull-forward with hindsight and as prudent stocking ahead of an N2 ramp in real time. It is the single cheapest early-warning indicator on this stock, and the next reading comes on 15 October.
“Our business in the second quarter was supported by strong demand for our leading-edge process technologies,” said Wendell Huang, Senior Vice President and Chief Financial Officer of TSMC. “Moving into third quarter 2026, we expect our business to be supported by continued strong demand for our leading-edge process technologies, including the steep ramp-up of our 2-nanometer technology.”
What TSMC’s customers and rivals are doing about it
The response that matters most came from Santa Clara. Nvidia reported revenue of $96.2 billion for the quarter ended 26 July 2026, up 106% year on year, with data-centre revenue of $89.0 billion, up 117%, and guided the current quarter to $108.0 billion plus or minus 2% — explicitly assuming no data-centre compute revenue from China, per its earnings release of 26 August 2026. Our coverage of that print noted the market’s ambivalence: the beat was enormous and the stock still fell on the margin guide.
Nvidia is not absorbing TSMC’s price increase quietly, either. It is passing cost downstream, having already raised AI server prices as memory costs climbed. That is the tell for the bull case: when the customer facing a wafer price rise responds by raising its own prices rather than resisting the increase, the pass-through chain is intact and TSMC’s rent is being paid by someone further down the line.
Apple sits at the other end. It has the least room to make an increase visible, because it either raises device prices, claws savings back elsewhere in the bill of materials, or eats the margin — a squeeze we set out in our Apple bull and bear analysis. AMD and Qualcomm sit between the two, with AMD better placed given its data-centre accelerator mix.
The competitive response has been to build, not to compete on price. Samsung Foundry continues to chase leading-edge share from a much smaller base, as covered in our Samsung bull and bear piece, while Intel’s foundry ambitions remain a capital story rather than a market-share story — see our Intel analysis. Neither has taken a visible bite out of TSMC’s advanced-node mix. Memory is the one place where a rival supplier has genuine pricing power of its own, which is why SK Hynix and Broadcom have re-rated alongside the logic names.
TSMC is also expanding without carrying the whole balance sheet. On 11 August 2026 it agreed a joint venture with Sony Semiconductor Solutions for next-generation image sensors in Koshi City, Kumamoto, with Sony contributing about ¥465 billion and TSMC about ¥282 billion, Sony as sole controlling shareholder and volume production planned for 2029, according to the announcement. A small deal in revenue terms and a large one in signalling: TSMC will take a minority seat where the technology is adjacent rather than core.
“AI has reached its inflection point. It’s doing useful work. Its tokens are productive and profitable. Now, compute is revenue,” said Jensen Huang, founder and CEO of NVIDIA, in the same 26 August release. “And demand is accelerating.”
The capex arithmetic, and what it has to earn
TSMC’s annual report set 2026 capital expenditure at US$52–56 billion, against US$40.9 billion actually spent in 2025. That budget was subsequently lifted to $60–64 billion, and on 11 August the board approved a further US$29.44 billion of appropriations for advanced capacity, advanced packaging and fab construction. The spend targets 2nm and 3nm capacity — Fab 20, Fab 21 in Arizona and Fab 22 — plus specialty and packaging at Fab 24 in Dresden.
Here is the synthesis that decides the stock. In Q2 2026 TSMC generated NT$783.36 billion of operating cash flow and spent NT$496.00 billion on capital expenditure, leaving free cash flow of NT$287.36 billion. Capital expenditure absorbed 63% of operating cash flow in a quarter with a record 67.7% gross margin. Run the same arithmetic at a 58% gross margin — TSMC’s level as recently as Q2 2025 — and free cash flow at that capex intensity goes to roughly nothing. The margin is not a bonus on top of the growth story; it is the thing paying for the growth story.
That is why the price increase and the capex increase arrived in the same quarter: they are the same decision. TSMC is funding a $60 billion-plus building programme out of current pricing, and the 2027 increase is what keeps the equation solvent once Arizona, Kumamoto and Dresden carry a full depreciation load. The overseas dilution is real, and the offset is being collected in advance, from customers, as list price.
| Driver | Bull reading | Bear reading |
|---|---|---|
| Revenue per wafer up ~14.6% YoY | Structural pricing power in a market with one leading-edge supplier | Half of 2026 growth is rent that can reset; volume alone grew 16.6% |
| Gross margin 67.7%, above guidance | Operating leverage on a fab base already built | Peak margin, and the comparison base for every quarter from here |
| Capex $60bn+ and rising | Demand visibility good enough to pre-commit a decade of capacity | Fixed cost that does not fall if orders soften; 63% of operating cash flow |
| Top ten customers 78% of revenue | Deep, contracted relationships with the strongest balance sheets on earth | Record concentration; roughly a dozen order books set the P&L |
| 2027 price rise of 5–10% | Margin defence already negotiated and banked | Signals that overseas cost dilution is real and large enough to need offsetting |
| Working capital: 87 inventory days, 29 A/R days | Prudent stocking into the N2 ramp | Up 11 and 6 days respectively; classic late-cycle signature |
Concentration deserves its own line. TSMC’s ten largest customers were 78% of net revenue in 2025, up from 76% and 70% in the two prior years. What is more revealing is the internal shuffle: the largest customer fell from 25% of revenue in 2023 to 19% in 2025, while the second largest rose from 11% to 17%. A falling top-customer share is usually read as diversification. Here it is the opposite — it is a baton pass from the smartphone franchise to the AI accelerator complex, happening while total top-ten concentration hit a record. TSMC has fewer effective counterparties than at any point in its listed history, not more.
Regulation, Taiwan and the currency
The regulatory position is unusually favourable and unusually fragile at the same time. TSMC Arizona signed agreements with the US Department of Commerce in November 2024 for up to US$6.6 billion in direct CHIPS Act funding and up to US$5 billion of proposed loans. Those incentives are what make Arizona economics tolerable rather than punitive, and they are a policy instrument, not a contract right in perpetuity. ESMC in Dresden has a parallel European arrangement.
The push-pull runs in both directions. Washington wants leading-edge logic on American soil badly enough to subsidise it; it also administers the export-control regime that decides which of TSMC’s customers may ship which accelerators to which jurisdictions. Nvidia’s decision to guide the October quarter assuming zero data-centre compute revenue from China is the clearest available measure of how much revenue that regime has already removed from the system — and TSMC sits one layer beneath every one of those decisions without a vote in any of them.
Currency is the underrated risk. Substantially all of TSMC’s sales are denominated in US dollars, while over half of its capital expenditure is in currencies other than the NT dollar and its cost base is heavily Taiwanese. Q3 2026 guidance of 65% to 67% gross margin is explicitly conditioned on NT$32 to the dollar; the second quarter averaged NT$31.60. A materially stronger NT dollar compresses reported margin without a single wafer changing hands, and no amount of leading-edge capability protects against it. Geography is the other unmodellable tail: most of TSMC’s capacity, and most of its suppliers’, sits on one island, and Arizona, Kumamoto and Dresden dilute that over a decade, not a quarter.
The $610 bull case
Start from consensus. Nineteen analysts model FY2026 revenue of NT$5.44 trillion, up 42.69%, and EPS of NT$107.64, up 62.48%, per stockanalysis.com. Each ADR represents five ordinary shares — TSMC’s own release converts NT$27.25 of quarterly EPS into US$4.31 per ADR at the quarter’s average rate — so NT$107.64 becomes about US$16.82 per ADR at NT$32. At $414 the stock trades on roughly 24.6 times this year’s expected earnings.
The bull case does not require a bubble multiple. It requires 2027 to grow more slowly than 2026 and margins to hold. Assume USD revenue growth decelerates from “slightly above 40%” this year to about 22% next, with gross margin held near 66% as the January price increase offsets overseas dilution, and net margin around 53%. That produces roughly $21 per ADR in 2027. Put 29 times on it — a modest re-rating from today’s 24.6 times, and below where the stock traded at its $479 high on this year’s numbers — and you get $610, about 47% above spot and comfortably inside the $440-to-$700 range the sell side already occupies.
The causal chain is specific. First, N2 ramps through 2027 at prices set before the capacity existed, repeating the N3 pattern. Second, the 5–10% list increase lands in January and shows up in the Q1 2027 margin line. Third, advanced packaging remains the binding constraint on AI accelerator shipments, which keeps allocation — and therefore pricing — in TSMC’s hands rather than its customers’. If all three hold, $610 is arithmetic rather than optimism.
The $265 bear case
The bear case is not a demand collapse. It is a composition problem and a multiple problem arriving together.
Assume 2027 USD revenue is flat on 2026 as hyperscalers digest what they have bought — not a decline, merely a pause. The price component of growth reverses first, because incremental HPC capacity above forecast is exactly where the 10–15% premium sits and exactly what gets cancelled. Gross margin falls toward 58%, TSMC’s level in Q2 2025, as newly built N2 capacity runs below plan and the overseas fabs carry full depreciation. Net margin drops to roughly 44%, and ADR earnings fall to about $14.50. Apply 18 times — a normal multiple for a cyclical capital-goods business with 63% of its operating cash flow committed — and the stock is $261. Round to $265, some 36% below spot and still 17% above the 52-week low of $225.63.
Watch the funding structure for the trigger. Nvidia announced partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilise over $500 billion of third-party capital for AI infrastructure — a figure its own release describes as subject to definitive agreements. When the marginal buyer of TSMC’s leading-edge output is a financing vehicle rather than a hyperscaler’s operating cash flow, the demand curve acquires an interest-rate sensitivity it did not previously have. That is the transmission mechanism from a credit event to a foundry order book, and it did not exist two years ago.
What we are watching next
Three dated checkpoints. On roughly 10 September, TSMC reports August revenue; the January–July run rate of NT$2,872.06 billion, up 37.0%, needs monthly prints above NT$450 billion to keep the full-year “slightly above 40%” guidance credible. On 15 October, Q3 results land against guidance of US$44.6–45.8 billion and a 65–67% gross margin — and the inventory and receivable day counts will say more than the headline. In January 2027 the price increase takes effect, and the Q1 2027 margin line will show whether it stuck at 5% or at 10%.
Our base expectation sits closer to the bull case than the bear, but not because demand looks infinite. It is because the pricing decision has already been negotiated, and pricing is the half of TSMC’s growth that consensus is not modelling explicitly. The risk to that view is not an AI winter. It is a strong New Taiwan dollar and a customer that decides its 2027 order was 15% too big.
Frequently asked questions
What is the TSMC stock prediction for 2027?
Our bull case is $610 and our bear case is $265, against a close of $414.00 on 1 September 2026. The bull case assumes about 22% USD revenue growth in 2027 with gross margin near 66%, valued at 29 times roughly $21 of ADR earnings. The bear case assumes flat revenue, margin reverting to 58%, and an 18 times multiple.
Why is TSMC’s gross margin so high?
Gross margin was 67.7% in Q2 2026, above the 65.5–67.5% guidance range. Advanced technologies at 7nm and below made up 77% of wafer revenue, and blended revenue per twelve-inch-equivalent wafer rose about 14.6% year on year. Leading-edge mix and price, not volume alone, are doing the work.
How concentrated is TSMC’s customer base?
Very. The ten largest customers accounted for 78% of net revenue in 2025, up from 76% in 2024 and 70% in 2023, per the company’s Form 20-F. The largest customer was 19% and the second largest 17%. Those ten also represented 84% of accounts receivable at the end of 2025.
How much is TSMC spending on capacity in 2026?
The annual report set 2026 capital expenditure at US$52–56 billion, later lifted to a $60–64 billion range, against US$40.9 billion spent in 2025. The board separately approved about US$29.44 billion of capital appropriations on 11 August 2026 for advanced capacity, packaging and fab construction.
What happens to TSMC prices in 2027?
TSMC has negotiated base increases of 5% to 10% effective from the start of 2027, with mature nodes including 12nm, 16nm and 28nm rising by as much as 10%, and a further 10% to 15% premium on high-performance computing capacity booked above a customer’s original forecast.
What would most damage the bull case?
A materially stronger New Taiwan dollar, because guidance is set at NT$32 to the US dollar and virtually all sales are dollar-denominated; and cancellation of above-forecast HPC bookings, because that is precisely where the premium pricing sits. Watch inventory days, which rose to 87 from 76 year on year.
This article is analysis, not investment advice. Prices and forecasts were accurate at the time of writing on 2 September 2026; TSM last closed at $414.00 on 1 September 2026 and traded at $409.66 in Tuesday’s pre-market session.
