TSMC posts $46.7 billion quarter and says AI chip demand still exceeds supply

TSMC reported record third-quarter results on October 15, 2026, and its chairman and CEO C.C. Wei said AI chip demand still outruns what the company can build. Revenue was $46.72 billion for the quarter ended September 30, above the top of the company's own forecast, and TSMC guided to $52.3 billion to $54.3 billion for the fourth quarter. For anyone tracking the physical side of the AI race, this is the clearest hard number of the month: the foundry at the center of AI chip supply says supply, not demand, is the limit.

The numbers from the quarter

TSMC's own results table shows third-quarter net revenue of $46.72 billion, a gross margin of 66.4% and an operating margin of 58.7%. In July the company had guided to $44.6 billion to $45.8 billion, so the quarter beat the top of that range by about $0.9 billion.

Coverage from CRN Asia, which reported the full release, adds that revenue rose 41.2% from a year earlier and 16.2% from the second quarter, that net income was NT$763.06 billion (about US$23.9 billion), up 68.7%, and that diluted earnings were NT$29.43 per share. Gross margin slipped 1.3 percentage points from the prior quarter, which is within the 65% to 67% range TSMC had guided.

The Motley Fool's write-up of the report, dated October 16, 2026, says high-performance computing, the platform that holds AI accelerators, rose 22% from the second quarter and now makes up 68% of revenue, up from 57% in the third quarter of 2025. That is the number that matters for this site: more than two thirds of the world's most advanced chip foundry is now selling into data-center compute and its neighbors.

What Wei actually said

On demand, Wei said on the earnings call, as quoted by the Motley Fool: "the demand actually is still far above our capability today to supply." He added that TSMC "did not see any sign of slowdown" in AI accelerator growth.

CRN Asia's account of the call is a paraphrase, not a transcript, so treat these as reported remarks. It says Wei described demand for AI accelerators, CPUs for agentic AI and networking chips as pushing silicon demand higher, said TSMC is producing advanced packages at five times the reticle size, and said customers still prefer CoWoS, TSMC's advanced packaging, as their first choice even though a competitor offers a strong alternative. It also says TSMC is planning cautiously for 2027 in consumer and price-sensitive segments because of rising component costs and macroeconomic uncertainty.

That last point is easy to miss. The boom is concentrated. Phones made up 21% of revenue, and TSMC is guarded about them. The AI line is where the company sees no slowdown.

The guidance, and what it implies

For the fourth quarter TSMC guided to revenue of $52.3 billion to $54.3 billion, a gross margin of 65% to 67% and an operating margin of 56.5% to 58.5%, using an exchange rate of NT$31.8 per dollar. CRN Asia calculates the midpoint at roughly 14% above the third quarter and 58% above a year earlier. It also reports full-year 2026 revenue growth of about 44% in US dollars, up from a little over 40% in July, and a 2026 capital budget of $60 billion to $64 billion, with 2nm capacity expected to more than double in 2027 and Arizona's second fab scheduled for volume production in the third quarter of 2027. Those last items come from a single secondary report, so check them against TSMC's management report before building on them.

Margin deserves a note. A 66.4% gross margin at this scale is extraordinary for a manufacturer. It still fell 1.3 percentage points from the prior quarter, per CRN Asia. Even a record quarter did not lift margin.

The catch: demand above supply is not the same as demand forever

A supply-constrained foundry is a good sign for chip buyers' appetite and a bad sign for anyone who wants cheaper compute soon. Capacity, packaging and power all gate how fast labs can scale. We covered the investor-side wobble in the eight-day AI chip stock losing streak and the geopolitical side in Nvidia's falling China share. TSMC's report speaks to neither of those directly. It tells you orders are still arriving faster than chips can ship.

There is also a limit to what one quarter shows. TSMC does not break out which customers or which labs drive the AI line, so these figures cannot tell you whether a specific lab's training plans are on track. A foundry beating guidance proves orders and shipments; it does not prove the models trained on those chips will pay back the spending.

How to read the 68% figure

The high-performance computing label is broader than AI accelerators alone, and the sources we opened do not split AI out. What they do show is direction: the segment's share of revenue went from 57% in the third quarter of 2025 to 68% in the third quarter of 2026, per the Motley Fool's reading of the report. Over twelve months, the company's mix has tilted sharply toward the customers building data centers, which is also why a slowdown in lab spending would show up here early.

Hype check

The claim is that AI demand has no sign of slowing. The evidence for the supplier side is strong: audited results, a guidance raise and a record quarter. The evidence that end-customers earn their money back is not in this report at all.

Our take

Read TSMC as the physical-race scoreboard. A $46.72 billion quarter and a $52.3 billion to $54.3 billion forecast say the buildout continues at full speed through at least year-end. We would watch two lines next: whether the AI-heavy HPC share keeps rising above 68%, and whether TSMC's cautious tone on 2027 consumer demand spreads to its AI guidance.