Built by Grok 4.6 High

Reports Thursday, October 15, 2026, at 14:00 Taiwan time (2:00 a.m. Eastern), before the NYSE open. Prepared September 24, 2026. This is the unreported Q3 2026 print (quarter ended September 30). The baseline is Q2 2026, reported July 16, 2026. The quiet period is October 5–14. September monthly sales are scheduled for October 8, inside that window and before the call.

TSM | Earnings Preview

Q3 2026 has not printed. The baseline is Q2 2026. USD revenue was $40,201 million, +33.7% YoY versus $30,070 million. That rate decelerated 690 bps from Q1 2026’s +40.6% ($35,898 million versus $25,526 million). The slowdown is the compare, not a demand break management believes: Q1 2025 was the soft quarter and Q2 2025 was already a much higher base. Profit grew more than twice as fast as revenue. Diluted EPS was NT$27.25, +77.4% YoY versus NT$15.36. ADR diluted EPS was $4.31, +74.5% YoY versus $2.47. Do not mix the two EPS series, or the two net-income series: parent net income was NT$706,562 million, +77.4%, and $22,359 million, +74.3% versus $12,825 million.

Gross margin was 67.7%, +910 bps YoY versus 58.6%. Operating margin was 60.3%, +1,070 bps YoY versus 49.6%. Margin expansion accelerated into Q2 (gross-margin YoY was +740 bps in Q1). The Q3 guide breaks that. HPC is why the top line stayed in the 30s while smartphone did not: HPC was 66% of revenue versus 60% a year earlier, and HPC dollars were NT$830.4 billion, +48% YoY. Smartphone was 22% versus 27%, down 500 bps of mix, with dollars only +13% YoY. 2nm showed up for the first time at 3% of wafer revenue. 3nm was 30% versus 24%. Wafer shipments were 4.336 million, +16.6% YoY, against revenue +33.7%. The gap is price and mix.

The guide still in force, spoken July 16 and not revised, is Q3 revenue $44.6 billion to $45.8 billion. The midpoint is $45.2 billion, which management called about +12% sequential or +37% YoY. On the reported Q3 2025 base of $33,097 million that midpoint is +36.6% YoY, about 290 bps faster than Q2’s +33.7% and still slower than Q3 2025’s own +40.8%. Gross margin 65% to 67%, midpoint 66%, is +650 bps YoY versus 59.5% and a deliberate 170 bp step down from the Q2 actual. Operating margin 56% to 58%, midpoint 57%, is +640 bps YoY versus 50.6% and about 330 bps below the Q2 actual. The FX assumption is $1 = NT$32. Full-year 2026 USD growth is “slightly above 40%,” the third raise this year (close to 30% in January, above 30% in April). Capex is $60 billion to $64 billion.

Why the year guide already implies Q4. FY2025 USD revenue was $122.43 billion, +35.9% versus $90.07 billion. Exactly +40% is about $171.4 billion. First-half 2026 was $76.1 billion. A Q3 midpoint of $45.2 billion leaves Q4 near $50.1 billion, about +48% YoY versus $33,731 million. “Slightly above 40%” makes Q4 larger unless Q3 beats the midpoint. That Q4 figure is arithmetic, not a guide.

What to watch

• The revenue bar the street is using is the midpoint, and the monthly tape is not. FMP’s October 15 estimate is $45.258 billion and $4.39 ADR EPS, inside the range and about $58 million (+0.1%) above the $45.2 billion midpoint. July consolidated revenue was NT$467.58 billion, +44.7% YoY, and August was a record NT$514.81 billion, +53.3% YoY (company releases, not a Daloopa quarterly print). At the company’s own FX of 32, September has to fall about 6% from August to avoid clearing the high end, and about 10% to land on the midpoint. A firmer realized NT dollar makes the same NT$ worth more dollars. Consensus has not marked that.
• The margin sandbag is thinner than the revenue sandbag. Gross margin cleared the prior high end by +200, +130, +120, then only +20 bps in Q2. The new caveat is a 3 to 4 point second-half gross-margin hit from the 2nm ramp, partly offset by utilization and cost. A print above 67% would mean that dilution did not show up. A print at 66% is what they asked for.
• Tariffs left the prepared remarks. Component prices did not. July dropped the tariff paragraph that had been in the remarks since January. Consumer and price-sensitive end markets are still “being challenged.” Section 232 treatment of Taiwanese chips is still a framework, not a Q3 revenue haircut. China was already 6% of Q2 revenue versus 9% a year ago.
• The call is the forward guide. Q3 dollars are largely knowable once September sales print on October 8. What is not knowable is the Q4 range, whether “slightly above 40%” moves again, whether 3nm passes 5nm (30% versus 33% last quarter), and whether C.C. Wei puts a number back on an AI-accelerator CAGR he called “stronger and stronger” and would not restate. Only Micron (September 30) and ASML (October 14) still print before this call. Neither is a TSMC wafer-revenue print.

The contradiction that matters: management has raised the in-year number twice, has a record of then roughly hitting the revised year, and says the five-year AI demand is stronger than the mid- to high-50s CAGR from January. The published street number is still the Q3 midpoint. They agree on the margin pause. They do not yet have a Q4 number to agree on.

The last print was a high-end revenue result and a margin beat, and the ADR still finished the July 16 session -2.32% and the next session -2.77% (Daloopa adjusted closes; no fundamental id on the price payload). Contemporaneous coverage framed that as the higher capex budget and a second-half gross-margin outlook below the Q2 actual, not as a miss of the Q2 guide. That is the reaction path into a call whose new information is the same two topics. No live price, market cap, or multiple is used here.

Issued July 16 on the Q2 call. No later release has revised the quarterly revenue range, the margin ranges, “slightly above 40%,” or the $60–64 billion capex band. Daloopa stores the forward quarter on the source quarter (2026Q2). Classification: revenue is conservative versus an eight-quarter habit of finishing above the midpoint, and in line with a street that has not followed July–August sales. Gross margin is only mildly conservative. Operating margin is the conservative line. The full-year phrase and the capex range are raises, not sandbags. EPS is not guided.

Over the last eight guided quarters, USD revenue finished above the midpoint every time, by an average of about +2.2%, and at or above the high end in six of eight. The last print landed on the high end: $40,201 million versus $40.2 billion. The dollar width of the guide has stayed near $0.8–1.2 billion while revenue compounded, so this range is only about 2.7% of the midpoint. A perfect “above midpoint” record with a shrinking range is a sandbag that is getting less useful. Gross margin never missed the low end. The last four beats versus the high end are +200, +130, +120, and +20 bps. Operating margin has not shown that fade: the last four beats versus the high end are +310, +300, +210, and +180 bps (Q2 actual 60.3% versus a guided high end of 58.5%). A pattern description, not a forecast, puts Q3 near $45.9 billion and about 67.2% gross margin, through both high ends. The 3 to 4 point 2nm dilution is why the margin habit may stop even if the revenue habit does not.

FMP’s Q2 2026 revenue actual of $39.356 billion does not match company revenue and is not used. Historical surprises below use Daloopa actuals against FMP estimates.

Q3 2026 is not a reported column. Platform rows are not guided. Internal is n/a on every row. YoY is versus the year-ago quarter, not the prior quarter. The sequential margin step is management’s own bridge and is labeled as such.

Current quarter — Q3 2026

The NT$ year-over-year rate on the translated guide is faster than the USD rate because Q3 2025 was booked at 29.91, not 32. That is currency, not a second demand view. TWD consensus of NT$1,436.8 billion (15 analysts) is the same call as the USD midpoint. Implied EBITDA margin uses FMP EBITDA of NT$1,060.9 billion over that revenue average.

FQ+1 — Q4 2026

Not guided. TSMC does not guide the following quarter until the print. Nothing on the July call was a Q4 revenue, margin, or EPS range. The USD shape below is the full-year phrase solved for Q4. It is a calculation.

If Q3 instead prints the high end ($45.8 billion), the same +40% year leaves Q4 near $49.5 billion, still about +47% YoY. A high-end Q3 beat, an unchanged 40% year, and a normal seasonal Q4 cannot all be true. Either Q4 really is that large, or the year phrase gets lifted on October 15 the way “around 30%” became “close to mid-30s” in October 2025. At NT$32, FMP’s NT$1,577 billion is about $49.3 billion, in the same place. TWD consensus YoY of +50.8% uses Q4 2025 NT$1,046,090 million.

FY+1 — FY2027, and the FY2026 guide it sits under

FY2027 is not guided. The live full-year guide is FY2026. FY2027 consensus is an unguided street path.

First-half 2026 USD growth was +36.9% ($76.1 billion versus $55.6 billion). Getting the year to slightly above 40% requires the second half to accelerate. The Q3 midpoint of +36.6% does not do that by itself. The acceleration is the unguided Q4. FY2026 implied margins use FMP EBIT of NT$3,223.5 billion and EBITDA of NT$4,235.9 billion over revenue of NT$5,373.2 billion. The through-cycle gross-margin floor of 56% and higher is not the binding constraint: Q2 already printed 67.7%.

Eight reported quarters

Columns are reported periods. Q3 2026 is omitted because it has not been reported. Growth is year over year.

Revenue growth troughed at +25.5% in Q4 2025 and reaccelerated to +40.6%, then cooled to +33.7% on a harder compare. The Q3 midpoint (+36.6%) extends the reacceleration versus that trough and does not require a new peak versus Q3 2025. Gross-margin YoY expansion did the opposite through Q2: it widened from +170 bps to +910 bps. The guide says that expansion continues at a slower pace. HPC mix is a five-year fact that is still making higher highs. 5nm is still the largest node. 2nm has no column yet except Q2 at 3%. Operating-margin YoY before Q3 2025 is left blank because the year-ago operating margin was not in the sourced set. Q2 capex of $15.7 billion was +63% YoY versus $9.63 billion.

Five-year annual trajectory

The down year was 2023. Since then USD growth has stepped from +30% to +36%, and the company is guiding an acceleration to slightly above +40%. That is still an acceleration year inside the same regime, not a new company. HPC taking share from smartphone is a five-year fact. What is new in 2026 is that the incremental node (2nm) dilutes margin while the stock of leading-edge revenue is still expanding it. FY2022 revenue YoY and the net-income YoY rates are arithmetic on the Daloopa levels in the table. FY2021 capex was not in the sourced annual set and is left blank.

The live setup is still the July 16 guide. Management has not revised it. Confidence is high that Q3 USD revenue lands inside or through the top of $44.6 billion–$45.8 billion, and that full-year 2026 USD growth finishes slightly above 40%. Confidence is lower that gross margin beats the range the way it has recently. Tone versus January and April is more bullish on AI demand and on the year, and more careful about annualizing customer orders. C.C. Wei would not promise an uninterrupted ramp through 2030.

What changed across the three calls. January was “close to 30%,” capex $52 billion–$56 billion, 2nm dilution of 2% to 3% for the full year, and an explicit tariff caveat. The AI-accelerator CAGR for 2024–2029 was raised to the mid- to high-50s, and the company CAGR to about 25%. April inched the year to “above 30%,” pushed capex to the top of the old range, and added an unquantified Middle East chemicals risk while saying the fabs would not be disrupted. Wei told the street to wait until July for a tighter growth number. July is the inflection: “slightly above 40%,” capex reset to $60 billion–$64 billion, conviction “very high,” and the 2nm drag resized to about 3 to 4 points in the second half. Tariffs dropped out of the prepared remarks. Component prices and the consumer caveat stayed, word for word enough that they have not once stopped a guide raise. Treat them as the residual risk statement, not as the forecast.

They have earned the right to be believed on a one-quarter revenue range. Q1 revenue of $35.9 billion was slightly above a $34.6–35.8 billion guide. Q2 revenue of $40.2 billion was the high end of a $39.0–40.2 billion guide. Gross margin beat the top of the prior range by about 130 bps, 120 bps, and then 20 bps. The pattern argues for the high end on revenue. It does not argue for another gross-margin beat of the same size, because July is the first quarter they guided the margin down to absorb 2nm. Full-year 2025 was guided up through the year and finished +35.9%, above the “close to mid-30s” language from the Q3 2025 call. The October call is where they have admitted the year.

Assumptions under the guide. Six profitability factors, with these called out for the second half. Very strong leading-edge demand, productivity, and across-node cost work are the offsets. In April they expected N3 gross margin to cross the corporate average in the second half. That crossover was not walked back. Overseas fabs stay a 2 to 3 point early drag, widening to 3 to 4 points later. That is not a Q3-specific surprise. The Q3 gross-margin range is struck at 32, not at the Q2 realized 31.60. On the October 2025 call a 1% move in the dollar versus the NT dollar was about 40 bps of gross margin. They have not restated that sensitivity. Use it as the last company figure, not as a new guide. Customer orders are discounted: Wei’s line is that every customer tells him the truth, and the sum of those truths is not the truth. Capacity is set with a top-down and bottom-up cut, including checks on data-center buildings so wafers do not land in inventory. The January AI CAGR was not expanded to include data-center CPU. Agentic-AI CPU demand is extra silicon. The additional $100 billion Arizona plan has no firm schedule and is not an add-on to the 2026 capex budget. The tax-rate band of 17%–18% was not updated in July. Q2’s roughly 20% tax rate was a one-quarter undistributed-earnings accrual.

What he refused. No 2027–2028 revenue number. No three-year capex number, beyond “even more significantly higher” than the prior three years. No updated AI CAGR. No date when supply and demand balance. No promise there is no dip before 2029–2030. He would not adopt an analyst’s 30–50% figure for how far unconstrained 3nm-and-below demand sits above supply. “The gap is very big” was as far as he went. Packaging is tight enough that he said it is limiting customer growth. He welcomed outside alternatives, including EMIB-T, because they would unclog front-end wafers. No separate packaging revenue number. Pricing stays strategic, not opportunistic. He said he would be happy around 68% gross margin and will not impose a multi-fold hike. Q2 was already 67.7%.

Post-guide evidence, July 16 through September 24. No financial guidance has been changed. Monthly revenue is the only hard datapoint, and it supports the USD range. Consolidated, unaudited, from TSMC’s monthly table and the August 6-K. September is blank.

That bridge uses the company’s own FX assumption of 32. It is not a forecast and not a TSMC USD disclosure of July or August. A September that repeats August is about $1.0 billion above the high end. Do not read the +45% to +53% NT$ months as a beat of the +37% USD guide by themselves. July plus August, NT$982 billion, is already 99% of the entire Q3 2025 NT$ quarter of NT$989,918 million, with September still out. January–August NT$ growth of +39.3% sits next to a full-year USD guide of slightly above 40%. Those are not the same currency. The +39.3% does not by itself force another full-year raise, and it does not miss the guide either.

FX has moved both ways inside the quarter. Early August closes were above 32. The Taipei interbank close on September 24 was 31.780, about 0.9% stronger than the assumption of 32, and September 23 was 31.716. A firmer NT dollar makes a given NT$ total worth more dollars, so the revenue-beat case gets larger, not smaller. It is a gross-margin headwind on the order of a few tenths of a point, inside a 200 bp range, using last year’s 40 bps per 1% rule of thumb. The quarter-average rate is not the September 24 spot. Do not haircut 65–67% for the spot alone. Applying 32 to months that were booked nearer 31.7 understates how far ahead the quarter already is in dollars. At about 31.7, July plus August is roughly $31 billion, and September has to fall about 9% from August’s dollar pace to finish at $45.8 billion.

Events since the call were technical, not financial. SEMICON Taiwan (September 2–4) did not update revenue, margin, or capex. A reported CoWoS capacity CAGR above 80% through 2027, attributed in trade press to a September 21 Kaohsiung event, is not a company guidance revision. The Baipu groundbreaking is a validation line and a training center, not 2026 volume. The September 23 North America OIP forum and a Synopsys certification of A14 design flows are consistent with the July statement that A14 is on track. A14 is not a 2026 revenue driver: pre-production 2027, volume production 2028. No CEO or CFO change. No date change. No guidance withdrawal.

Ranked by what can still change the October 15 debate. Q3 revenue is mostly a September sales print. The call is the margin, the mix, and the forward guide. “What the street has” means FMP as of September 24 plus the company’s own guide. There is no Visible Alpha node or platform model.

DCE is the only platform in outright year-over-year decline in Q2 (NT$12.7 billion, -15.6%). IoT was NT$64.1 billion, +44.6%, and auto was NT$53.7 billion, +24.7%. Both are too small to move a $45 billion quarter. Wei rejected a broad mature-node recovery. The shortage he described is power-management and sensors tied to AI data centers. Inventory days rose, which management attributed to the N2 ramp, not to a demand air pocket. That comment is from the call, not a separate Daloopa series pulled for this page.

Window is July 17 through September 24, after the Q2 report and before the Q3 print. The July 16 release itself is the baseline, not a row. Nothing below is a Q3 result. Nothing revises the 65–67% gross-margin range. SharePoint, OneNote, Outlook, and internal models were not available. Rows were kept only where a primary release or a named wire (Reuters, Bloomberg, Nikkei, TSMC, ASML, or the Central Bank) stood behind them.

Left out on purpose: the July 16 capex raise and the $100 billion Arizona plan, which are the baseline; price moves and 13F or Form 4 activity; supply-chain wafer-start figures (2nm from about 90,000 wafers a month to 110,000, CoWoS doubling) that TSMC has not confirmed; and a US award to GlobalFoundries that was not confirmed by Reuters, Bloomberg, or Commerce. A14 construction commentary from the Central Taiwan Science Park is a building schedule, not a 2026 wafer guide.

What matters for October 15 is narrower than the headline list. The quarter is not being reset by a guide change, a customer loss, or a quantified quake. Two unaudited months are running faster than the USD guide. The 2027 price report is the fundamental offset to 2nm and overseas dilution, and management has not confirmed the percentage. Samsung–Broadcom is a crack in a book that has had no leading-edge alternative, and it is still an envelope, not a named migration of the AI accelerator book.

Consistent beater on ADR EPS: 12 of 12 and 4 of 4. Revenue is mixed over 12 quarters (10 of 12) and a consistent beater over the last 4 (4 of 4), but the revenue beat is shrinking. Both revenue misses were under half a percent (Q4 2023 -0.26%, Q2 2025 -0.47%), and in both of those quarters the company still landed inside or above its own USD revenue guide. EPS beat magnitude is larger recently (last-four average +9.0% versus +5.4% over the prior eight) and widened again in Q2 to +11.4%. Revenue beat magnitude has narrowed for four straight prints, from +3.2% in Q3 2025 to +0.9% in Q2 2026. The street is close on revenue and still low on EPS. That split is the margin.

Surprise is (Daloopa actual − FMP estimate) / FMP estimate. Actuals are not FMP’s recorded actuals. Q2 2026 FMP revenue actual of $39.356 billion does not match $40,201 million. Q4 2024 FMP EPS actual of $2.19 does not match $2.24; using the FMP actual would have flipped that quarter from a beat to a miss. Those FMP actuals are not used. A beat is any positive surprise. No quarter was a dead zero. Q3 2026 is not on the grid.

Green is a large beat (#1E8449: revenue at or above +3%, EPS at or above +7%, or a guide-mid beat at or above +3%). Blue is a mid-size beat (#1A5276). Yellow is a small beat (#D4AC0D: revenue under +1.5%, EPS under +3%, or a guide-mid beat under +2%). Red (#C0392B) is a miss versus FMP. The guide-mid row starts at Q3 2024 because that is the window with a full prior-call pair in the sourced set. Versus the company’s own guide, revenue was never below the low end in the twelve quarters, and gross margin was never below the guided low end.

The repeating explanation is not a demand miss. When management discusses a variance, it is utilization and cost running ahead of the three-month-old forecast, with the exchange rate as the swing factor and overseas fabs as the offset. The one quarter they missed FMP revenue by a still-sub-1% amount, Q2 2025, they were above their own guide. The NT dollar had appreciated versus the assumption, which management said cut NT$ revenue and gross margin. The street was already above that guide, so a guide beat was still a small consensus miss.

What that sets up for October 15: FMP revenue of $45.258 billion sits near the midpoint of $44.6 billion–$45.8 billion, not above the high end the way the Q2 2025 estimate did. A print at the high end would be about a 1% beat of FMP. A print at the low end would be about a 1.5% miss. The revenue-surprise trend says the street has closed the gap. The EPS history says the opposite on the bottom line. Consensus of $4.39 is only about 2% above the Q2 print of $4.31, while the last four EPS beats averaged 9.0%. The offset is the margin guide. They have cleared the gross-margin high end in four of the last four quarters, usually on utilization, cost, and a better exchange rate than the one they assumed. That is the pattern. It is not a Q3 result.