Built by Grok 4.6 HighTaiwan Semiconductor reported third-quarter 2026 results on Thursday, October 15, 2026. This review was prepared Monday, October 26, 2026. Quarter ended September 30, 2026. Calendar fiscal year, so the slug is 2026Q3. The earnings calendar labels October 15 as 2026Q4; the call, the filing, and Daloopa all say this print is the third quarter. Next print is the fourth quarter, dated January 21, 2027 on the FMP file. SharePoint, OneNote, Outlook, and Excel were not available. Consensus is FMP as of October 26, not Visible Alpha or Bloomberg.
TSM | Earnings Review — Q3 2026
The top line sped up. The margin rate did not. USD revenue of $46,720 million grew +41.2% versus $33,097 million. That is +747 bps of acceleration from Q2’s +33.7%, and it is still 325 bps short of the Q2 2025 peak at +44.4%. The split inside the 41% is the new fact: wafer shipments of 4.476 million grew only +9.6%, while implied ASP grew +28.8%. A year ago wafers were the faster leg. HPC is 68% of revenue, from 57%. 2nm is 8% of wafer revenue, from 3% in Q2. Smartphone dollars are about flat.
Gross margin is still expanding a lot versus last year, and the expansion narrowed. Gross margin of 66.4% is +690 bp versus 59.5%. Q2’s expansion was +910 bp. Operating margin of 58.7% is +810 bp, down from +1,070 bp. NT$ diluted EPS of NT$29.43 grew +68.8%, −866 bps from the +77.4% peak. ADR EPS of $4.60 grew +57.5%. The gap between those two EPS rates is the weaker NT dollar, not a second earnings quarter. USD/NTD averaged 31.98 versus 29.91.
Double beat, and the two surprises moved in opposite directions. Versus the October 26 FMP file, revenue was +$1,053 million / +2.3% and ADR EPS was +$0.15 / +3.4%. Revenue also cleared the July high end of $45.8 billion by $920 million, the widest dollar clearance of the guide high in this window. Gross margin landed inside 65%–67%, 60 bp under the high, after four quarters above it. That is why a wider revenue beat became a smaller EPS beat. The record is still 12/12 on ADR EPS and 10/12 on revenue. Do not use FMP’s stored revenue actual of $46.060 billion. The filing is $46.72 billion, and $46.7 / $40.2 is the +16.2% sequential rate the CFO said. $46.06 billion is not.
New guide: a faster quarter, a fourth raise of the year, and no new capex number. Q4 revenue is $52.3 billion to $54.3 billion. The midpoint, $53.3 billion, is about +58% versus Q4 2025 revenue of $33.73 billion. Full-year 2026 USD growth moved from “slightly above 40%” to around 44%. Capex stays $60 billion to $64 billion. FMP’s Q4 revenue estimate is $52.579 billion, inside the range and near the floor. Gross-margin guidance of 65–67% holds the level and cuts the year-over-year expansion to about +370 bp.
What the street has not underwritten, and the one live break. Wei left the AI-accelerator CAGR at mid-to-high 50s for 2024–2029 and said it “probably will be higher.” CPUs stay outside that KPI. N2-family capacity is expected to more than double in 2027. FMP’s 2027 NT$ year is about +38% versus its own 2026 year, a deceleration. The disclosure break is overseas dilution: prepared remarks put 3–4 points “in the next few years,” and the Q&A put that same 3–4 points back in the latter part of the five-year window. Arizona volume production in Q3 2027 is the front of the April “second half of 2027” window, not a pull-in past it.
| USD revenue | $46,720M · +41.2% YoY · +747 bps · +2.3% vs FMP | NT$ revenue | NT$1,494,243M · +50.9% YoY · +1,490 bps · FX, not a second business |
| Wafers / implied ASP | 4.476M wafers · +9.6% · ASP +28.8% · 1.096 x 1.288 = 1.412 | HPC / smartphone | 68% / 21% of revenue · HPC dollars about +68% · smartphone about flat |
| 2nm / 3nm | 8% / 33% of wafer revenue · 2nm was 3% in Q2 · advanced 7nm-and-below 79% | Gross margin | 66.4% · +690 bp YoY · expansion narrowed from +910 bp · inside 65–67% |
| Operating margin | 58.7% · +810 bp YoY · +70 bp above the 58% guide high | NT$ / ADR EPS | NT$29.43 · +68.8% · ADR $4.60 · +57.5% · +3.4% vs FMP |
| EBITDA (calc.) | $34,287M · +54.2% YoY · margin 73.4% · +620 bp · off the 76.0% Q2 rate | Q4 guide vs FMP | Mid $53.3B / +58% vs $52.579B / +55.9% · GM mid 66% |
| FY2026 | Around 44% USD · Q4 mid implies about $176.1B and +43.9% · capex $60–64B unchanged | Beat record | ADR EPS 12/12 · revenue 10/12 · last four quarters 4/4 on both |
The foundry sells wafers. The growth rate is how many, times what they sell for, times who buys them. Q3 followed price and HPC, not unit growth. Read the trajectory in USD. NT$ growth of +50.9% is the same quarter translated at a weaker NT dollar: 1.412 times 1.069 is 1.509.
| Metric | Q4'24 | Q1'25 | Q2'25 | Q3'25 | Q4'25 | Q1'26 | Q2'26 | Q3'26 |
|---|---|---|---|---|---|---|---|---|
| USD revenue ($M) | 26,884 | 25,526 | 30,070 | 33,097 | 33,731 | 35,898 | 40,201 | 46,720 |
| USD revenue YoY | +37.0% | +35.3% | +44.4% | +40.8% | +25.5% | +40.6% | +33.7% | +41.2% |
| Rev accel (bps) | +96 | −174 | +916 | −360 | −1,535 | +1,516 | −694 | +747 |
| NT$ revenue YoY | +38.8% | +41.6% | +38.6% | +30.3% | +20.5% | +35.1% | +36.0% | +50.9% |
| Wafers, 12-inch (M) | 3.418 | 3.259 | 3.718 | 4.085 | 3.961 | 4.174 | 4.336 | 4.476 |
| Wafer YoY | +15.6% | +7.6% | +19.0% | +22.4% | +15.9% | +28.1% | +16.6% | +9.6% |
| Implied ASP YoY | +18.5% | +25.8% | +21.4% | +15.1% | +8.3% | +9.8% | +14.6% | +28.8% |
| HPC mix | 53% | 59% | 60% | 57% | 55% | 61% | 66% | 68% |
| 3nm mix | 26% | 22% | 24% | 23% | 28% | 25% | 30% | 33% |
| 2nm mix | — | — | — | — | — | — | 3% | 8% |
| Gross margin | 59.0% | 58.8% | 58.6% | 59.5% | 62.3% | 66.2% | 67.7% | 66.4% |
| GM YoY (bp) | +600 | +570 | +540 | +170 | +330 | +740 | +910 | +690 |
| Operating margin | 49.0% | 48.5% | 49.6% | 50.6% | 54.0% | 58.1% | 60.3% | 58.7% |
| EBITDA ($M, calc.) | 18,445 | 17,709 | 20,984 | 22,236 | 23,446 | 26,096 | 30,541 | 34,287 |
| EBITDA YoY | +43.1% | +36.3% | +50.3% | +35.9% | +27.1% | +47.4% | +45.5% | +54.2% |
| Diluted EPS (NT$) | 14.45 | 13.94 | 15.36 | 17.44 | 19.50 | 22.08 | 27.25 | 29.43 |
| EPS YoY | +56.9% | +60.2% | +60.7% | +39.1% | +34.9% | +58.4% | +77.4% | +68.8% |
| EPS accel (bps) | +284 | +334 | +44 | −2,159 | −413 | +2,345 | +1,902 | −866 |
| ADR EPS ($) | 2.24 | 2.12 | 2.47 | 2.92 | 3.14 | 3.49 | 4.31 | 4.60 |
Last five wafer YoY prints: +22.4%, +15.9%, +28.1%, +16.6%, +9.6%. Last five ASP YoY prints: +15.1%, +8.3%, +9.8%, +14.6%, +28.8%. Volume times price equals the revenue rate. Implied HPC dollars are about $31.8 billion, roughly +68%. North America is 79% versus 76%. China is 5% versus 8%. One platform and one geography set the growth rate.
EBITDA dollars accelerated to +54.2% because revenue did. The margin stepped off 76.0% to 73.4%. Q3 operating income was $27,403 million and D&A was $6,884 million.
Five-year shape. FY2023 was the down year. FY2024 and FY2025 were a speeding upcycle at the year level: USD revenue +30.0% then +35.9%. Around 44% for 2026 would be a third year of faster growth. HPC mix went from 37% to 58% over the five years and is already 68% in the latest quarter. Gross margin, not EBITDA margin, is the profit-rate story, because depreciation rose with the fabs.
| Fiscal year | FY2021 | FY2022 | FY2023 | FY2024 | FY2025 |
|---|---|---|---|---|---|
| USD revenue ($B) | 56.83 | 75.89 | 69.30 | 90.07 | 122.43 |
| USD revenue YoY | +24.9% | +33.5% | −8.7% | +30.0% | +35.9% |
| HPC mix | 37% | 41% | 43% | 51% | 58% |
| Gross margin (calc.) | 51.6% | 59.6% | 54.4% | 56.1% | 59.9% |
| GM YoY (bp, calc.) | −147 | +793 | −520 | +176 | +377 |
| Diluted EPS (NT$) | 23.01 | 39.20 | 32.34 | 45.25 | 66.24 |
| EPS YoY | +15.2% | +70.4% | −17.5% | +39.9% | +46.4% |
Consistent beater on the ADR line. Revenue surprise re-widened. EPS surprise compressed. A beat is any positive surprise versus the October 26 FMP earnings file. Actuals are Daloopa, not FMP’s stored actuals. Using FMP’s $46.060 billion would shrink this revenue beat from +2.3% to about +0.9%. The September 24 pre-print bar was $45.258 billion and $4.39. This quarter is a beat on either bar. The post-print file is the smaller surprise, and it is the one used so every quarter sits on one pull.
| Surprise vs FMP | Q4'24 | Q1'25 | Q2'25 | Q3'25 | Q4'25 | Q1'26 | Q2'26 | Q3'26 |
|---|---|---|---|---|---|---|---|---|
| Revenue | +2.5% | +0.3% | −0.5% | +3.2% | +2.2% | +1.5% | +0.9% | +2.3% |
| ADR EPS | +1.8% | +2.4% | +3.8% | +11.0% | +8.3% | +5.4% | +13.1% | +3.4% |
| Revenue vs guide | In | In | Above high | Above high | Above high | Above high | At high | Above high |
| GM vs guide | At high | In | In | Above high | Above high | Above high | Above high | In range |
Over twelve quarters, revenue beat 10/12 and ADR EPS beat 12/12. The two revenue misses were 2023 Q4 (−0.3%) and 2025 Q2 (−0.5%). Last four quarters are 4/4 on both lines. Average revenue surprise rose from +1.4% in the prior eight to +1.7% in the last four. Average EPS surprise rose from +5.4% to +7.6% only because Q4 2025 and Q2 2026 were large. This print’s +3.4% is the second-smallest EPS beat in twelve quarters.
Management did not say “versus the street.” They said the quarter was above their own USD guidance on leading-edge demand, and that gross margin fell 1.3 points sequentially primarily due to 2nm ramp dilution. FX was on the assumption: the guide used NT$32.0 and the quarter averaged 31.98. Currency explains the NT$-versus-USD growth gap. It does not explain the USD beat. No one-time item was offered.
NT$ revenue of NT$1,494,243 million was about +3.0% versus the separate FMP TWD average. NT$ EPS per ADR of NT$147.13 was about +3.9% versus that file’s NT$141.57. Those are cross-checks, not the twelve-quarter score.
TSMC guides one quarter ahead and a full-year growth phrase. It does not guide EPS. The phrase has moved one way: close to 30%, above 30%, slightly above 40%, around 44%. Capex moved once, in July, to $60–64 billion, and was not moved again. Revenue has been at or above the high end for four straight quarters. Gross margin cleared its high by 20 to 130 bp from Q4 2025 through Q2 2026. Q3 2025 is the fourth above-high quarter in the twelve-quarter record. Q3 2026 landed inside.
Waterfall, Q4 dollars. There was no prior Q4 range. The $50.1 billion bar is the residual of a flat 40% year after the old Q3 midpoint. It is not a guide. The new midpoint is about $3.2 billion above that residual. The Q3 beat was $1.52 billion above the old midpoint, so most of the year raise is a higher Q4, not the quarter already printed.
| Q4 2026 | Guide low | Guide mid | Guide high | FMP |
|---|---|---|---|---|
| Revenue ($B) | 52.3 | 53.3 | 54.3 | 52.579 |
| Revenue YoY vs $33.73B | +55.1% | +58.0% | +61.0% | +55.9% |
| Gross margin | 65% | 66% | 67% | n/a |
| GM YoY vs 62.3% | +270 bp | +370 bp | +470 bp | n/a |
| Operating margin | 56.5% | 57.5% | 58.5% | n/a |
| OM YoY vs 54.0% | +250 bp | +350 bp | +450 bp | n/a |
| ADR EPS | Not guided | Not guided | Not guided | $5.16 |
| FX, NT$ per USD | — | 31.8 | — | — |
FMP is about $0.72 billion / 1.4% under the midpoint and about 210 bp slower than the company’s 58%. The TWD consensus of NT$1,669.8 billion is about $52.5 billion at the 31.8 assumption. Same place: inside, near the floor. Eleven days after the call, the published book had not moved to the midpoint.
Nine-month revenue on the billions series is about $122.82 billion. Add the Q4 midpoint and the year is about $176.1 billion, +43.9% versus $122.43 billion. The low end of the Q4 range implies +43.0%. The high end implies +44.7%. “Around 44%” is the top half of the range, not the floor. Plugging FMP’s $52.579 billion into the year-to-date produces about $175.4 billion, +43.3%. The street year sits near the bottom of what the company just said.
Capex of $60 billion to $64 billion is unchanged. Nine-month capex is $11.10 billion + $15.70 billion + $13.84 billion = $40.6 billion. That leaves $19.4 to $23.4 billion for Q4, versus Q4 2025 capex of $11.51 billion. Reiterating the budget is not restraint. Q4 has to be the heavy tool quarter or the low end is missed. There is no 2027 capex dollar. The phrase “even more significantly higher” than the prior three years, about $101 billion, was repeated and not sized.
FY2027 is qualitative only. FMP’s NT$7,570 billion is about +38% versus its own 2026 average. That is a deceleration the company did not endorse. N2 capacity more than doubling is a capacity statement, not a revenue CAGR.
The tone shift versus July: they raised the year and did not raise the capital budget. They had room, after a print above the revenue high, to lift spending again. They did not. The margin caveat is the same 2nm ramp, now visible in the print rather than only in the warning.
USD revenue YoY is three acts, not a new peak. A mid-30s grind, a Q2 2025 local high at +44.4%, a slide to +25.5% in Q4 2025, a snap-back, a Q2 2026 give-back, and this quarter’s +747 bp re-acceleration to +41.2%. NT$ revenue, the operating series, troughed in the same quarter and has accelerated three times, to a window-high +50.9%. EPS did not follow. It peaked at +77.4% and cooled 866 bps to +68.8%. The dollar EPS add still widened, from +NT$11.89 to +NT$11.99. The base rose faster than the add. That is slower margin expansion plus a tougher compare, not an earnings-dollar rollover.
The entire USD-versus-NT$ gap this quarter is translation. FX YoY was +6.9% more NT$ per dollar. In Q3 2025 the currency ran the other way, so USD growth sat well above NT$ growth. This quarter laps that strong-TWD base. A model that treats +50.9% as a faster wafer business is double-counting the currency. Management guides in USD.
Inflection points worth keeping:
- Q4 2025 was the air pocket, not a new trend. USD revenue YoY fell 1,535 bps to +25.5%. Both series bottomed there.
- Q2 2026 was the margin-rate peak and the EPS-rate peak, not the revenue-rate peak. Gross-margin expansion +910 bp. EPS +77.4%. Revenue YoY gave back 694 bps.
- Q3 2026 splits the two. Revenue re-accelerates. EPS and the pace of margin expansion do not. The Q4 guide asks investors to underwrite a still-faster revenue rate and a still-slower margin expansion, about +370 bp at the gross-margin midpoint.
The January 21, 2027 print tests the quarter that makes the 44% year. The number that can still change the out-year is the one Wei would not print.
| Catalyst | When | What the street has | What management added |
|---|---|---|---|
| Q4 and the 44% year | 2027-01-21 | FMP $52.579B, low end of $52.3–54.3B. A low-end print matches the street and misses the sentence. | Leading edge, including the 2nm ramp, supports Q4. Around 44% USD, from slightly above 40%. |
| 2nm versus gross margin | Q4 mix, then 2027 | No FMP gross-margin series. The guide mid is 66%, 40 bp under Q3. | 2nm went from 3% to 8% of wafer revenue. Q3 dilution versus Q2 was 1.3 points, inside the July 3–4 point warning. N2 family capacity more than doubles in 2027. |
| AI accelerator CAGR | Refused | Still mid-to-high 50s, 2024–2029, narrow definition. 2027 NT$ consensus is a slowdown to about +38%. | Probably higher. No slowdown seen. CPUs are not in the number. The supply gap is still quite big. |
| 2026 capex delivery | Q4 spend | No 2027 capex estimate. The only number is the reiterated $60–64B. | About $19–23B has to land in Q4. 2026–28 still significantly higher than the prior three years. Revenue is supposed to outpace capex later, not in 2026 itself. |
| Consumer versus N2 handsets | Already in the mix | The +38% 2027 year is where the caution has to sit. No platform split in FMP. | Price-sensitive end markets stay challenged into 2027. The same remarks say N2 capacity doubles on smartphone and HPC together. |
| Price on the signed book | Asked, not answered | No percent in FMP or on the call. | Pricing is strategic and differs by geography. He did not confirm that reported increases reopen the contracted book. |
He has earned the near-term guide. Q3 revenue was above the high end, operating margin was above the high end, and 2nm dilution came in milder than the July 3-to-4-point gross warning. The thing he keeps repeating, and the street has not put in the model, is the out-year supply gap. A Q4 print at $52.6 billion with another refusal to touch the CAGR is exactly the quarter the current consensus describes. It would not test the upside.
Two prints would say the caution, not the conviction, was the forecast. Revenue at or below $52.3 billion, with HPC mix no longer rising. Or gross margin below 65% while 2nm is still a small share of wafers. Neither is what the guide says.
Well answered on the operating rules. Deflected on the quantities that reset the model. Seventeen scored items: 11 well answered, 6 deflected. The deflections cluster on the AI CAGR, wafer-capacity growth, a named IDM’s outsourcing, which Arizona phase sits under the dilution math, accelerator-versus-CPU quantities, and whether price hikes reprice the book already signed. The dodges were evasive-positive. The direction on the CAGR was up. The revision was not owned.
| Analyst | Asked | Answer that matters | Badge |
|---|---|---|---|
| Laura Chen, Citi | How capacity is set against power, funding, HBM, packaging | Those are already in the plan. Demand is still above supply. | Well answered |
| Laura Chen, Citi | Update the mid-to-high-50s AI CAGR | It does not matter. It probably will be higher. No new number. | Deflected |
| Gokul Hariharan, JPMorgan | Can wafer growth reach about 10%. Is 2027 tighter. | No percent. Customers keep raising demand, so he would not call 2027 tighter or looser. | Deflected |
| Gokul Hariharan, JPMorgan | Texas and a customer Terafab | Evaluation and a discussion. Any fab has to be run by TSMC. He did not say the customer’s name. | Well answered |
| Krish Sankar, TD Cowen | Customers passing wafer allocations | No subletting. Mask licensing is very small. | Well answered |
| Charlie Chan, Morgan Stanley | IDM insourcing of server CPU at A14 | No comment on one customer. Internal capacity is already in the plan. A14 is a bigger node than N2 without assuming that outsourcing. | Deflected |
| Charlie Chan, Morgan Stanley | If EMIB-T cannot deliver, will TSMC back the customer up | Yes. CoWoS is still first choice. 5x reticle, he said, is TSMC today. Working on 9x and 14x. | Well answered |
| Sunny Lin, UBS | Does 3–4 points of overseas dilution start in 2027 | That 3–4 points is still the latter part of the five-year period. He did not say 2027. | Deflected |
| Arthur Lai, Macquarie | Does the AI CAGR include CPUs | No. They cannot separate an AI server CPU from a conventional one. CPU demand is still part of the business. | Well answered |
| Sahej Singh, Stifel | Do reported price increases hit the signed book | Pricing depends on segment, location, and technology. He did not confirm a hike or a reopening. | Deflected |
| Sahej Singh, Stifel | GlobalFoundries interposers | A back-end shortage. The deal lets both support TSMC’s customer. Not a front-end concession. | Well answered |
Two other answers belong in the model as rules, not as numbers. High-NA is still a 2030 tool, and being later than a competitor is not, on Wei’s account, a technology gap. There is no ceiling on profitability, and there will not be a dramatic increase overnight. Power checks in the U.S. have not cut a customer forecast, and they have not pulled customers onto Super Power Rail.
Two statements on this print cannot both be the 2027 margin bridge. One date claim overstates a tightening that did not leave the prior window. One vendor actual cannot be the quarter the company reported. A later guide that management labels as a raise is not a contradiction. The currency word “TWD” stamped on the Arizona $265 billion total is a transcript defect, not a new capex number. The live capex guide remains $60–64 billion.
Prepared remarks, same call. Wendell Huang said dilution from overseas fabs is expected to widen to 3%–4% in the next few years, because more U.S. fabs will ramp. The clause that puts the near-term figure “closer to 2%” is grammatically broken in this transcript and is not a new point estimate.
Q&A, same call. Asked whether that wider dilution starts in 2027, he said the latter part of the five-year period is still 3 to 4 points, “and that still remain the case.” July’s prepared remarks were the same split: 2%–3% early, 3%–4% in the latter stages. The five-year window was anchored in January 2025 as starting that year.
What to model. Early-stage overseas dilution stays 2%–3% until the company names the year the 3%–4% starts. Do not model a 2027 step-up off Sunny Lin’s question, and do not replace the early band with 2% off the broken sentence.
April 16, 2026. C.C. Wei: the second Arizona fab, on 3nm, starts volume production in the second half of 2027. Construction was already complete.
October 15, 2026. The same fab starts volume production in the third quarter of 2027, “slightly ahead of our previous schedule.”
What to use. Q3 2027 is July through September. That is inside the second half. A date inside the prior window is not ahead of the window. Use Q3 2027. Do not take another quarter off the April schedule. The construction status does not conflict.
The call and the filing. Revenue increased 16.2% sequentially to $46.7 billion. Daloopa income-statement revenue is $46,720 million. The selected billions line is $46.72 billion. Prior-quarter revenue of $40.2 billion makes +16.2% only off $46.7 billion.
The file. FMP’s revenueActual for October 15 is $46.060 billion, which is about +14.6% sequential, not what the CFO said. FMP’s $4.60 EPS actual matches the ADR and is usable. The revenue actual is not. Q2 in the same file does match the company.
Explicitly not breaks. The move from “slightly above 40%” to “around 44%” is a labeled raise. Capex reiterated at $60–64 billion is not a cut. A 3–4 point gross 2nm hit in the second half, partly offset, can produce the 1.3 point sequential gross-margin decline that printed. The AI CAGR was not officially revised. “Prudent into 2027” and “extremely robust” leading edge are split on purpose, between consumer and the leading edge. Long-term gross margin of “53% and higher” is not a ceiling that 66.4% violates.
The macro on this call is a barbell. Price-sensitive and consumer end markets stay challenged into 2027. AI accelerator demand is still above what they can ship, and the narrow CAGR is probably above the mid-to-high 50s. Nothing was said about rates, tariffs, or export controls. Do not invent those.
| Who | Direction | The fact from this print |
|---|---|---|
| HPC and accelerator customers | Better, still short | HPC is 68% of revenue, from 57%. Implied dollars about +68% YoY. Demand above supply after they have already haircut for power, funding, and components. U.S. build checks were “so far so good.” |
| Server CPU | Extra, unquantified | Agentic-AI CPU demand is rising and is explicitly outside the accelerator CAGR. Do not mark the mid-to-high-50s number up because of CPUs. |
| Price-sensitive handsets and consumer | Worse into 2027 | Smartphone mix 21% versus 30%. Implied smartphone dollars about flat. The caution sentence now runs into 2027. Flagship N2 demand is the offset, not the weak tier. |
| ASML | 2030–31, not 2026 | High-NA use starts in 2030. A 12-inch photomask with ASML is a 2031 item. Being later than a competitor is not, on this call, a disadvantage. Near-term tool supply is tight and tool-price inflation is real, on an unnamed supplier set. |
| Other wafer-fab equipment | Better, no new dollar | 2026 capex reiterated at $60–64B after the July raise. 2026–28 stays significantly higher than the prior three years. Most of the dollar is still the demand plan, not inflation. |
| GlobalFoundries | Better, back-end only | Wei confirmed an interposer deal because the back end is a little short. It supports TSMC’s customers. It is not a front-end share shift. It does not add 2026 CoWoS wafers. |
| Intel packaging | Mixed | EMIB-T was the analyst’s name. Management said the competitor’s solution is competitive and that TSMC will be the backup if it does not deliver. Not a statement that a program has failed. July had welcomed outside packaging taking load. October still needs a partner and now offers to catch a failure. |
| China revenue | Not the growth | Mix 5% versus 8%. Management did not discuss China policy. The fact is only that the growth is not there. |
Q3 platform dollars are not a filed series. The “about +68%” HPC figure and the “about flat” smartphone figure multiply whole-percent mix by cited revenue. They are ranges, labeled as calculations. The direction does not depend on the rounding: an 11-point HPC mix gap is not a rounding artifact, and smartphone cannot be growing with the company at a 21% mix versus 30%.