The easy question about the revenue TSMC filed this morning is whether August was a record. It was, and the question leads nowhere. Six of the eight months TSMC has filed this year set an all-time high, so "record" describes the trend line rather than telling you anything new about it — and the company's own release never uses the word. Answer it and you are holding a superlative.
The harder question is what a filed month does to a promise that is still open. Taiwan requires every listed company to publish monthly revenue by the tenth of the following month. That rule has now put two of the three months of TSMC's September quarter permanently on the record, while the guidance management issued for that quarter in July stands unchanged. Two fixed numbers against one fixed range mean the third month is no longer free. It has a requirement.
So the question here is narrow: what does September have to do for TSMC's guidance to survive contact with TSMC's own filings? It takes one conversion, one subtraction and five years of the company's September history — and the escape hatch everyone reaches for turns out to be pushing the wrong way.
What is now fixed
On 10 September TSMC reported August revenue of NT$514.81 billion, up 10.1% on July and 53.3% on August 2025. July, filed a month earlier on Form 6-K, was NT$467.58 billion. Together they are NT$982.4 billion, and neither figure can be revised.
Against that sits the guidance TSMC issued on 16 July alongside second-quarter results: third-quarter revenue of US$44.6–45.8 billion, gross margin of 65–67%, on a stated assumption of US$1 = NT$32. We covered that quarter and that guide in August. Nothing about it has been revised since.
One conversion and one subtraction
The guide is denominated in dollars; the filings are in NT dollars. Converting at TSMC's own stated assumption is the step that turns a record month into information.
At NT$32, US$44.6–45.8 billion is NT$1,427.2 billion to NT$1,465.6 billion. Subtract the NT$982.4 billion already on the record and what September must contribute falls straight out.
NT$ billions and month-on-month change from August 2026 — R40 arithmetic on TSMC filings
| Target or path | September revenue | vs August |
|---|---|---|
| Bottom of guide | NT$444.8bn | -13.6% |
| Midpoint of guide | NT$464.0bn | -9.9% |
| Top of guide | NT$483.2bn | -6.1% |
| Worst Sept in five years | NT$491.5bn | -4.5% |
| Last September | NT$507.5bn | -1.4% |
The first three rows are requirements: TSMC's US$44.6-45.8bn third-quarter guide converted at the company's own stated US$1 = NT$32 assumption (NT$1,427.2bn-NT$1,465.6bn), less the NT$982.4bn of July and August already filed. The last two rows are outcomes, applying the month-on-month change of the worst and the most recent of the last five Septembers to August 2026 actual revenue. All five are measured against August 2026 of NT$514.81bn.
To land at the top of its own guided range, TSMC needs September revenue to fall 6.1% from August. To reach the midpoint it needs a 9.9% fall. To touch the bottom, 13.6%.
Five Septembers
Those are large requirements for a month that has not behaved that way once in five years. TSMC publishes its own monthly history, and the last five Septembers moved like this:
- 2021: NT$137.4bn → NT$152.7bn, +11.1%
- 2022: NT$218.1bn → NT$208.2bn, −4.5% — the worst of the five
- 2023: NT$188.7bn → NT$180.4bn, −4.4%
- 2024: NT$250.9bn → NT$251.9bn, +0.4%
- 2025: NT$335.8bn → NT$331.0bn, −1.4%
Not one of them fell far enough. Repeat the worst September in five years and the quarter lands at about NT$1,473.9 billion, or US$46.06 billion at NT$32 — US$0.26 billion above the ceiling. Repeat last September and it lands near US$46.56 billion, US$0.76 billion above. A flat month puts it at US$46.79 billion.
That is the finding, and it is a disclosure fact rather than a forecast. With twenty days left in the quarter, TSMC's own guidance has stopped being a range and become a floor. Getting back inside it now requires September to be worse than any of the last five.
The escape hatch runs the wrong way
There is one honest way out, and it is not demand. The guide is in dollars; the revenue is earned in NT dollars. The rate between them is the only variable still genuinely open, because TSMC books its dollar revenue at realised rates rather than at the 32 it assumed in July.
The direction is easy to get backwards. A weaker NT dollar — a rate above 32 — shrinks the dollar print of any given NT$ month. A stronger NT dollar inflates it.
US$ billions at NT$32, and the NT$ per US$ average required to print US$45.8bn - R40 arithmetic
| September path | Q3 at NT$32 | Rate needed for US$45.8bn |
|---|---|---|
| Worst in five years, -4.5% | US$46.06bn | 32.18 |
| Last year, -1.4% | US$46.56bn | 32.53 |
| Flat | US$46.79bn | 32.69 |
Each row applies that September path to August 2026 actual revenue, adds the NT$982.4bn already filed, and converts. Column two uses TSMC's own guidance assumption of US$1 = NT$32. Column three is the quarterly average rate at which the same NT$ total would print exactly at the US$45.8bn top of guidance - a higher number means a weaker NT dollar. For reference, TSMC realised 31.60 in the June quarter and Bank of Taiwan's spot buying rate was 31.50 on 10 September. TSMC's realised rate reflects its own recognition timing and hedging and is not known until it reports on 15 October.
So the rescue needs the NT dollar to weaken to a quarterly average near 32.2 on the worst-September path, or 32.5 on last year's. It is not doing that. TSMC realised 31.60 in the June quarter — its own NT$1,270.38 billion against US$40.20 billion. Bank of Taiwan's spot buying rate on 10 September was 31.50, and has sat between 31.44 and 31.71 every business day this month. Two of the quarter's three months of currency are behind us as well: ASE Technology Holding's own paired NT$ and US$ figures for the same months imply about 31.95 in July and 32.18 in August.
On that path the quarter's average lands at or below the 32 assumption, which makes the dollar figure slightly larger, not smaller. The currency is widening the gap it would have to close.
One caveat carries real weight. TSMC's realised rate is its own convention — it turns on when revenue is recognised and on hedging nobody outside the company can see — so the third-quarter average is unknowable until it reports on 15 October. The breakeven rates above are firm; where the realised rate lands is not.
What it is actually worth
Very little, and saying so is the point. An overshoot of US$0.26–0.76 billion, at the guided 65–67% gross margin, is roughly US$170–500 million of gross profit — against a quarter that should produce close to US$30 billion of it, in a company that closed at $435.36 on 9 September, near US$2.26 trillion of market value on the diluted count. That is a rounding difference in the P&L.
It is also not a demand story, and nothing here licenses one. TSMC's monthly filing is a single consolidated line: no end-market, no node, no customer. The AI attribution in this morning's coverage is the reporters', not the company's. Nor is the back end confirming an acceleration — ASE Technology Holding's assembly, testing and materials business grew 7.9% month on month in August, slower than TSMC's 10.1%. The 11.5% figure that circulated is the holding company including its electronics-manufacturing arm, not the packaging business.
The fair objection is that TSMC guides conservatively and clearing the top of its range is routine, so "the guide is a floor" describes the company's guidance habits rather than the September quarter. That is right, and it is the honest size of the claim. The arithmetic does not establish that demand accelerated. It establishes that TSMC's published range no longer describes the quarter it was written for, and that anyone still treating US$45.8 billion as the upper bound is using a number the filings have overtaken.
Two things would change that. A September falling more than 6.1% would put the quarter back inside the range, and none of the last five has come close; that one settles on 10 October, when September revenue posts under the same rule that made this morning's number public. An NT dollar weakening past roughly 32.2 on a quarterly average would do it through the conversion instead, and that one waits for 15 October, when TSMC reports and its realised rate becomes visible for the first time.
August revenue is from TSMC's release of 10 September and its own monthly revenue tables, which also supply the 2021–2025 September history; that figure had not yet appeared on Form 6-K at the time of writing, and 2026 monthly figures are unaudited. July revenue, the second-quarter results and the third-quarter guidance are from TSMC's Forms 6-K of 10 August and 16 July. ASE Technology Holding's August figures are from its release of 9 September. Exchange rates are Bank of Taiwan spot buying rates on the dates stated; the implied July and August rates are derived from ASE's paired NT$ and US$ disclosures, and TSMC's June-quarter rate from its own. Calling August a record, and attributing it to AI demand, are press characterisations — TSMC's release does neither, and it publishes no US dollar figure for August. Every NT$ conversion of the guide, required September figure, breakeven rate and gross-profit and market-value estimate here is ours. The share price is the 9 September close.
