This round of decline is being treated as AI hardware peaking in the third quarter. If TSMC's (TSM) October revenue, announced on November 10, 2026, is still rising quarter-on-quarter, wouldn't that show that the third quarter was not the peak?

This round of decline is being treated as AI hardware peaking in the third quarter. If TSMC's (TSM) October revenue, announced on November 10, 2026, is still rising quarter-on-quarter, wouldn't that show that the third quarter was not the peak?

The premise of this is invalid: October 2026 monthly revenue will not be disclosed until November 10; today is November 1, and that document does not yet exist. TSMC's latest monthly revenue is the September data submitted on October 8: consolidated revenue of NT$5,118.57 yi, down 0.6% quarter-on-quarter and up 54.6% year-on-year. Last October's revenue was submitted on November 10, 2025; at this pace, this year's will probably also come around the same day.

Even if the November 10 document comes out and quarter-on-quarter growth continues, it would not prove that the third quarter was not the peak. At TSMC's third-quarter earnings call on October 15, the company already fixed its fourth-quarter guidance: revenue of US$52.3 yi to US$54.3 yi, with the midpoint 14% higher than the third quarter's actual US$46.72 yi. On the revenue level, the third quarter was never the peak; this was settled half a month ago, and there is no need to wait another month.

What could truly peak is two other things: quarter-on-quarter growth and gross margin. The signals these two give in the third-quarter report are exactly opposite: one has already turned, the other is still accelerating.

The word "peak" has three different meanings here: absolute revenue, quarter-on-quarter growth, and gross margin. Which one the market means determines whether it is right. Mixing them together means the argument can never be settled.

Revenue is accelerating; gross margin peaked in the second quarter

Third-quarter standalone: US dollar revenue 467.2 yi, up 16.2% quarter-on-quarter and up 41.2% year-on-year; NT dollar revenue 1 trillion 4,942 yi, up 50.9% year-on-year; gross margin 66.4%, operating margin 58.7%, net margin 51.1%; earnings per share NT$29.43, equivalent to ADR US$4.60. Both revenue and operating margin exceeded the upper end of guidance, while gross margin fell within the range.

Looking across the four quarters:

The shape of the gross margin line is like this:

The second quarter was 67.7%, the third quarter 66.4%, and the fourth-quarter guidance midpoint is 66.0%. The high point was in the second quarter, not the third. So the statement "peak in the third quarter" gets the timing wrong if it refers to gross margin, and gets the direction wrong if it refers to revenue.

Revenue quarter-on-quarter growth was 6.4%, 12.0%, and 16.2%, accelerating all the way; over the same period, gross margin was 66.2%, 67.7%, and 66.4%, surging and then coming back. Revenue was US$6.5 yi higher in one quarter, yet gross margin fell 1.3 points instead. That is the price of the 2nm ramp, and the bill is recorded in cost of revenue.

The peak in quarter-on-quarter growth was drawn by the company itself in guidance

The fourth-quarter guidance midpoint of US$53.3 yi implies +14.1% versus the third quarter's 467.2 yi; the low end of 52.3 yi is +11.9%, and the high end of 54.3 yi is +16.2%. The first pullback in quarter-on-quarter growth falls in the fourth quarter; the third quarter's 16.2% is the quarter-on-quarter high of this cycle. I accept this point: if "peak in the third quarter" refers to quarter-on-quarter growth, guidance does support it.

Year-on-year is another direction. The first quarter was +40.6%, the second quarter +33.7%, the third quarter +41.2%; for the fourth quarter, using the guidance midpoint against last year's same-period 337.3 yi, it is +58%. Year-on-year is moving up, not down.

Adding the four quarters together: 359.0 yi, 402.0 yi, 467.2 yi, plus the fourth-quarter midpoint of 533 yi, gives full-year 2026 revenue of 1,761 yi US dollars. The four quarters of 2025 were 255.3 yi, 300.7 yi, 331.0 yi, and 337.3 yi, totaling 1,224 yi. 1,761 versus 1,224 is +43.9%, which exactly matches the company's statement at the earnings call of "about 44% US dollar revenue growth in 2026."

How this 44% came about is also worth a look: on April 16, with the first-quarter report, the company said "above 30%"; on July 16, with the second-quarter report, it said "slightly above 40%"; on October 15, with the third-quarter report, it said "about 44%." Two upward revisions in half a year. To explain, under the premise that "the third quarter is the peak," a company raising its full-year target to 44%, the cost is very high.

October's quarter-on-quarter figure is the least informative number of the year

A positive October quarter-on-quarter figure is this company's seasonal action, not new information:

For three consecutive years, October quarter-on-quarter growth has been double-digit positive. Even if this October rises 10%, it is only "the same as in past years." Using a month that should rise seasonally to prove the trend is not broken proves nothing at all.

September's -0.6% is worth a mention, though: from 5,148.06 yi in August to 5,118.57 yi in September, it was the second quarter-on-quarter decline in 2026 apart from the February Spring Festival month; the first was -1.1% in April. Reading this as "peak" is the handiest reading, but in the same document, the first three quarters cumulative were 3 trillion 8,987 yi, up 41.1% year-on-year.

What matters for October is not positive or negative, but whether it is enough. Fourth-quarter guidance midpoint of US$53.3 yi, at an exchange rate of 31.8, converts to NT$1 trillion 6,949 yi, requiring a monthly average of 5,650 yi over three months. September actual was 5,119 yi. Three scenarios:

• October achieved 5,544 yi (quarter-on-quarter +8.3%): the midpoint can be achieved at a smooth pace, and the remaining two months only need to average around 5,650 yi per month
• October only 5,200 yi (quarter-on-quarter +1.6%): November and December each need to do 5,875 yi, and each month still needs to rise another 13%
• October less than 5,000 yi (quarter-on-quarter -2.3%): the following two months each need to do 5,975 yi, rising 19% in one month, basically impossible to achieve, and the guidance will have to be revised down

This is how the document dated November 10 should be used. It is not a criterion for "whether the third quarter is the peak"; it is a thermometer for "whether the fourth-quarter guidance can hold."

Transmission chain: revenue peak depends on orders, gross margin peak depends on depreciation

Working backward from the end: cloud companies' capital expenditure determines orders for GPUs and self-developed chips, orders determine TSMC's wafer starts and CoWoS advanced packaging occupancy, and further upstream are HBM (Micron), equipment (Lam, ASML), and packaging (Amkor, ASE).

This chain has two completely different landing points in the financial statements. The revenue peak lands on the "net sales" line, and the leading indicators are orders and lead times; the gross margin peak lands on depreciation and labor within "operating costs," and the leading indicators are process mix and capacity utilization. The two things can move in opposite directions at the same time, and that is what happened in the third quarter.

Look at the three numbers in the third-quarter brief together: HPC accounted for 68% of revenue (66% in the second quarter), HPC quarter-on-quarter +22% (+20% in the second quarter); 2nm accounted for 8% of wafer revenue (3% in the second quarter), adding 5 percentage points in one quarter; during the same period gross margin -1.3 points. The revenue mix is tilting toward AI and advanced processes, and gross margin is being eaten by ramp-up.

Unit price is also moving up, not relying on price cuts to exchange for volume. Quarterly dollar revenue of 467.2 yi divided by 12-inch equivalent shipments of 447.6 wan wafers gives about 10,437 dollars per wafer; the second quarter was 9,271 dollars, quarter-on-quarter +12.6%. Shipments increased only 3.2%, while revenue increased 16.2%; the difference is price and mix. In this calculation revenue includes advanced packaging while shipments count only wafers, so the absolute level will be on the high side; look at the direction.

Inventory turnover days 82 days, second quarter 87 days, same period last year 74 days, declining quarter-on-quarter. Accounts receivable turnover days 28 days, second quarter 29 days, same period last year 25 days. Neither has moved in a bad direction; at least from the statements, there is no channel stuffing and no customers dragging on payment.

What shows attitude more than inventory is spending. Capital expenditure in the first three quarters was 1兆2,900 yi New Taiwan dollars, about 407 yi US dollars at 31.7, while the full-year 2026 capital budget was set in January at 520 yi to 560 yi US dollars. That is to say, the fourth quarter still needs to invest 113 yi to 153 yi US dollars; the second quarter was 157 yi and the third quarter 139 yi, the same order of magnitude. Equipment is still being bought as usual; this is the company using its own money to bet that "demand has not peaked."

The next-quarter guidance of these four companies: none is moving down

First align the timeline: TSMC and Amkor are on calendar quarters, Lam is on a fiscal quarter ending September 27, and Micron is on a fiscal quarter ending September 3. The four "next quarters" are not the same calendar interval; you can only look at direction, not add them directly.

The direction is very clear: memory's next-quarter guidance +13.4%, equipment +4.6%, foundry +14.1%. The only one giving negative growth is Amkor at -2.1%, and its own explanation is that communications weakened as expected, computing and automotive industrial set records, and gross margin rose from 18.3% to 18% to 19%. What fell was handset volume, not the price of AI.

To call the third quarter the peak of AI hardware, one would have to explain the next-quarter guidance of all four companies at the same time, and the cost of that is not small.

The strongest sentence from the opposing side

A smart person will not argue with you about October revenue; he will attack from three directions.

First cut monthly revenue: that is the rearview mirror. CoWoS and advanced packaging have long delivery cycles, and the revenue recognized in October corresponds to orders placed two or three quarters earlier. A month-on-month rise in October only proves that orders already placed are being delivered, not that new orders are still coming. What is truly forward-looking is customer capital expenditure and prepayments; monthly revenue ranks behind these two.

Then cut profit: what the stock is killing is not revenue, but net profit. Third-quarter revenue quarter-on-quarter +17.6%, net profit only +8.0%, and net margin fell from 55.6% to 51.1%.

Here it should be made clear that the main reason net profit lagged is not gross margin, but non-operating income falling from 959 yi to 346 yi, a decrease of 63.9%; this block is most likely foreign-exchange valuation gains and losses and is not operating deterioration.

But the conclusion is unchanged: revenue rose 17.6% while profit rose only 8%, showing that the company's current profit elasticity is not as good as its revenue elasticity. And 2nm accounts for only 8%, and will still climb in 2027; overseas fabs have just started production, and both of these weigh on operating costs. If revenue rises further, for every 1 point gross margin falls, at the third quarter's 16% tax rate, quarterly net profit would be about 1.6% less.

The heaviest cut is not in the statements: even if the third quarter is not the revenue peak, that does not mean this decline is wrong. If the previous price was giving "40% compound growth every year," then with quarter-on-quarter growth falling to 14% and gross margin turning down, the valuation should be revised down. This is a valuation-level matter, and it can hold at the same time as whether revenue has peaked, and it does not need October data.

I accept these three points. They point to the same judgment: the third quarter is not the revenue peak, but it may indeed be the peak of "growth and gross margin."

If this number comes out, this judgment is wrong

• October revenue below 5,200 yi New Taiwan dollars: "revenue has not peaked" must be downgraded to "held up by the last two months."
• October revenue below 5,000 yi: the fourth-quarter midpoint requires November and December to each rise 19%, and the risk of a guidance cut becomes real.
• Fourth-quarter actual gross margin below 65% (the lower bound of guidance): this shows the share eaten by the 2nm ramp plus exchange rates is higher than the company itself estimated, and "gross margin is only mildly pulling back" is wrong.
• At the January 2027 earnings call, 2027 capital expenditure guidance below 2026's 52 billion to 56 billion US dollars, or 2027 US-dollar revenue growth guidance below 20%: only then has demand truly turned.
• Micron's fiscal 2027 second quarter, Lam's March 2027 quarter, Amkor's first quarter of 2027 — if any one of them actually turns negative quarter-on-quarter: the others turn first, and TSMC will follow.

In one sentence: the document on November 10 can answer "whether the fourth-quarter guidance can be held," but cannot answer "whether the third quarter was the peak." The latter has had an answer since October 15 — revenue is not the peak, gross margin is.

Source: Taiwan Semiconductor Manufacturing Company Limited Form 6-K monthly revenue announcements tsm-revenue20261008.htm (2026-10-08, 0001046179-26-000680), tsm-revenue20251110.htm (2025-11-10, 0001046179-25-000124), tsm-revenue20251009.htm (2025-10-09, 0001046179-25-000113), tsm-revenue20241108.htm (2024-11-08, 0001046179-24-000122), tsm-revenue20241009.htm (2024-10-09, 0001046179-24-000114), tsm-revenue20231110x6k.htm (2023-11-13, 0001046179-23-000019), tsm-revenue2023106x6k.htm (2023-10-06, 0001046179-23-000012); TSMC Form 6-K earnings releases and earnings call presentations EX-99.1 a3q26e_withguidancexfinal.htm, EX-99.2 a3q26presentatione.htm (2026-10-15, 0001046179-26-000726), a2q26e_withguidancexfinal.htm (2026-07-16, 0001046179-26-000451), a1q26e_withguidancexfinal.htm (2026-04-16, 0001046179-26-000199), a4q25e_withguidancexfinal.htm (2026-01-15, 0001046179-26-000008); Lam Research Corporation Form 8-K EX-99.1 lrcx_exhibitx991xq1x2027.htm (2026-10-21, 0000707549-26-000052); Amkor Technology, Inc. Form 8-K EX-99.1 amkr9302026erex-991.htm (2026-10-26, 0001047127-26-000049); Micron Technology, Inc. Form 8-K EX-99.1 a2026q4ex991-pressrelease.htm (2026-09-30, 0000723125-26-000018).