TSMC Q3 2026: Record NT$763 Billion Profit, 2nm Already 8% of Wafer Revenue

TSMC just posted its sixth consecutive record quarter, and the driver is the one every AI builder already feels: demand for AI chips. For the quarter ended September 30, 2026, the company reported net profit of about NT$763.1 billion (roughly US$23.5 billion), up nearly 69% from a year earlier, on revenue of about NT$1.49 trillion, up around 50%, according to Reuters-based reporting on Investing.com and CNBC. Analysts polled by LSEG had expected about NT$746.3 billion.

Update (October 15, 2026): Q4 guidance, Arizona timing and the demand check

We have now been able to read CNBC's report on the earnings call, which fills the gaps flagged below:

• Q4 guidance: TSMC forecast fourth-quarter revenue of US$52.3 billion to US$54.3 billion, about 14% higher than Q3 and about 58% higher than a year earlier at the midpoint, per CNBC. It said the quarter is supported by strong demand for advanced nodes, including the 2nm ramp.
• Capex: the company expects capital spending from 2026 to 2028 to be significantly higher than in the previous three years. CNBC did not give a revised 2026 figure, so the US$60 to 64 billion range cited later in this post remains TrendForce's number.
• Demand versus supply: SemiAnalysis analyst Sravan Kundojjala told CNBC that TSMC is checking whether customers have the financing and supporting components to deploy AI systems, and that "even after those checks, TSMC says demand remains well above its supply capacity." That is the analyst's account of the call, not a verbatim TSMC quote, and it speaks to the overcapacity worry some investors raised.
• Demand drivers: TSMC pointed to agentic AI and open-source AI models as expanding demand for advanced chips. High-performance computing was 68% of revenue and smartphones 21%.
• Arizona: the second fab is built and being equipped, with mass production now expected in Q3 2027, slightly ahead of schedule, according to CNBC.

This post separates what those reports confirm from what we could not verify. The headline numbers are consistent across outlets. The details beyond them, especially fourth-quarter guidance and the capex update, were not in the sources we could open, and we say so rather than guess.

Update (October 15, 2026): Q4 guidance and capex are now reported

Since this post went up, outlets citing TSMC's results have reported the outlook we could not open earlier. According to TokenPost, TSMC projected fourth-quarter revenue of US$52.3 billion to US$54.3 billion, with a gross margin of 65% to 67% and an operating margin of 56.5% to 58.8%.

For the full year, TokenPost reports TSMC raised its 2026 revenue growth outlook to about 44% in US dollar terms, from slightly above 40% previously. Seoul Economic Daily, citing Reuters, says TSMC raised its 2026 capital expenditure guidance to US$60 billion to US$64 billion and attributes the Q3 gross margin dip to 66.4% to costs of the early 2-nanometer ramp-up.

What this changes: the "Q4 guidance and capex" rows below that we marked as not confirmed are now reported by outlets, though we have still not read TSMC's own release, and the Q4 and growth figures rest on a single outlet. The quarter's dollar revenue also differs by outlet (roughly US$46.4 billion to US$46.7 billion depending on the exchange rate used). Verify against TSMC's investor relations page before relying on exact figures.

TL;DR: the TSMC Q3 2026 numbers

Why do the dollar figures differ between reports?

You will see revenue quoted as about US$46.36 billion in some summaries and about US$46.7 billion in earlier previews. The underlying figure, roughly NT$1.49 trillion, is the same. The gap is the exchange rate: TSMC reports in Taiwan dollars, and each outlet converts at its own rate, whether the quarter's average or the day's spot.

The same applies to profit, where the dollar figure is quoted anywhere from about US$23.4 billion to US$23.7 billion. When you cite TSMC, prefer the Taiwan dollar number and say which conversion you used. The percentages (up roughly 50% on revenue, up nearly 69% on profit) are unaffected.

How did the results compare with TSMC's own guidance?

In July, TSMC guided the third quarter to revenue of US$44.6 to 45.8 billion, a gross margin of 65 to 67% and an operating margin of 56 to 58%, as listed on TSMC's investor relations quarterly results page. The reported 66.4% gross margin sits inside that range, and the 58.7% operating margin is above the top of it. Revenue was reported above the guided range, per TrendForce.

That matters because the margin story had been a worry. In July, TrendForce reported that TSMC lifted its 2026 capex to US$60 to 64 billion and expected a Q3 margin dip, with the ramp of 2nm chips diluting second-half gross margin. Q3 came in without a visible margin hit, which suggests the dilution was either smaller than feared or offset by pricing and mix.

What does the 2nm and 3nm mix tell us?

Stacked chip layers illustrating the advanced 2nm and 3nm packaging behind TSMC AI accelerators

The most useful detail for builders is the node split. Per the Investing.com report, 3-nanometer chips were 33% of wafer revenue and 2-nanometer chips 8%. If those two figures are right, the leading edge accounts for about 41% of wafer revenue in a quarter when 2nm had only just started shipping in volume.

Why does that matter outside finance? Because the chips AI labs and cloud providers compete for are built on exactly these nodes. Custom accelerators like the ones we covered in Microsoft's Maia 300 inference chip and Meta's MTIA roadmap depend on the same foundry capacity as Nvidia and AMD parts. A fast 2nm ramp suggests the next generation of accelerators will have a manufacturing path, but it also means more customers bidding for the same wafers.

Where does this fit in the AI supply chain story?

Trays carrying a chip along a track, a metaphor for the AI chip supply chain after TSMC's record Q3 2026

TSMC is one link in a chain that is tight at several points at once. Our recent coverage maps the pieces:

• Foundry output: our weekly AI chip supply chain roundup tracked TSMC's record monthly sales, Samsung's phone cut and GlobalFoundries' interposers.
• Lithography: ASML's Q3 2026 results said AI demand exceeds EUV supply, with orders stretching toward 2029 and 2030.
• Optical parts: Lumentum reported optical components sold out toward 2029, another bottleneck for data centers.
• Packaging and US capacity: GlobalFoundries' Malta expansion is about silicon photonics and domestic fabs.
• Sovereign chips: Fujitsu's MONAKA CPU is fabricated by TSMC, a reminder of how many national programs still route through one foundry.

Read together, these say the same thing from different angles: demand for AI compute is not the constraint, manufacturing capacity is.

Why did the stock fall on a record quarter?

Investing.com noted that the results came amid "a deep rout in chipmaking and AI equity valuations," and its page showed TSMC's U.S.-listed shares down about 3.4%, with Nvidia down about 3.2%. The page warned that quote data may not be real time, so treat the exact moves as indicative.

The pattern is familiar: when expectations are already high, a beat is priced in, and attention shifts to guidance and capex. Pre-release previews cited by Reuters-based coverage said investors would focus on the fourth-quarter outlook and any change to the US$60 to 64 billion capex range. We could not retrieve those details, so we do not report them. If guidance disappointed, that, rather than the quarter itself, would explain a falling share price. That is an inference, not something the sources state.

What does this mean for what you build or pay?

This is a finance story, so here is the practitioner takeaway, stated carefully.

1. Expect continued capacity tightness. A foundry posting 66% gross margins and record profit is not a foundry with spare capacity. Cloud GPU and accelerator availability is likely to stay constrained, which tends to keep inference prices from falling as fast as model efficiency alone would suggest.
2. Plan around efficiency, not just supply. If next-generation hardware arrives slowly and expensively, techniques that cut tokens per task, such as caching, routing and smaller models, remain the dependable lever. Our piece on Microsoft's token-cost work shows what that looks like in practice.
3. Watch the 2nm ramp. If 2nm keeps growing from 8% of wafer revenue, newer accelerators with better performance per watt should follow. Performance per watt matters because power, not chips, is increasingly the limit for data centers.
4. Do not over-read one quarter. Profit is a lagging indicator. For product decisions, the leading indicators are cloud provider capacity announcements, GPU lead times and price lists.

What is claimed, and what is verified?

Being explicit about the evidence:

The headline Techmeme summary quoted revenue growth as 50.9% and net income growth as 68.7%. Investing.com's description of "nearly 69%" for profit and "up 50%" for revenue is consistent with that, with the difference being rounding. For anything you plan to cite formally, go to the source: the full release and management discussion are on TSMC investor relations.

Is this the peak, or is there more room?

A sixth consecutive record says the cycle has not turned, but it does not say how long it will last. Two things cut in opposite directions. On one side, capex of US$60 to 64 billion, raised from US$52 to 56 billion in July per TrendForce, shows TSMC is spending as if demand persists for years, and ASML's order book stretching toward 2029 points the same way. On the other side, the rout in chip equities mentioned by Investing.com shows investors are asking whether AI spending by customers can keep compounding.

For a builder, the sensible reading is that supply, not demand, sets the pace for now. Treat any forecast of a sudden compute glut with caution until orders, not just share prices, change. We will update this post if TSMC's guidance changes that reading.

How to check this yourself

• Open TSMC's quarterly results page and download the management report and presentation for 3Q26.
• Compare the stated gross and operating margins with the July guidance ranges above.
• Look for the platform split (HPC versus smartphone) in the presentation; HPC is the AI-driven segment.
• Read the Q4 guidance and the capex language in the transcript before drawing conclusions about next quarter.

What to watch next

• Q4 guidance and the capex range. The US$60 to 64 billion range set in July is the number to track.
• 2nm customer announcements. Which accelerator makers disclose 2nm products will indicate how much of the 8% is AI.
• Packaging capacity. Advanced packaging has been a repeated choke point, and any update on capacity is more informative for GPU supply than the revenue line.
• Export-control and geopolitical headlines, which can change where capacity can be sold regardless of demand.

Related reading

• TSMC and Musk's Terafab: the "operational control" claim vs. what is verified
• AI chip supply chain this week: TSMC record, Samsung phone cut, GlobalFoundries interposers
• ASML Q3 2026: AI demand exceeds EUV supply
• Lumentum optical parts sold out to 2029
• GlobalFoundries Malta, Japan and silicon photonics
• Fujitsu MONAKA: a Made-in-Japan AI CPU fabricated by TSMC
• Microsoft Maia 300 inference chip

Figures are as reported on October 15, 2026 and may be revised in TSMC's full filings. Dollar conversions vary by outlet.