TSMC reported third-quarter revenue of $46.72 billion on Thursday, October 15, up 41.2% from a year ago, and kept more than half of it as net profit.
If you want to know who is reliably making money from the AI boom, skip the model launches and look at the company that manufactures the chips they run on. These numbers read less like a chipmaker and more like a software company with a very expensive hobby.
It was TSMC’s sixth straight quarter of record net profit, and both revenue and profit beat analysts’ estimates, CNBC reported.
The numbers
For the quarter ended September 30, TSMC reported:
- Revenue: NT$1,494.24 billion, up 50.9% year over year and 17.6% from Q2. In dollars, that’s $46.72 billion, up 41.2% from a year earlier.
- Net income: NT$763.06 billion, up 68.7% year over year.
- Diluted EPS: NT$29.43, or US$4.60 per ADR unit.
- Margins: 66.4% gross, 58.7% operating, 51.1% net.
- Free cash flow: NT$476.56 billion, up from NT$139.38 billion a year earlier, per CNBC.
Against LSEG SmartEstimates, which weight the analysts with the best track records, revenue came in at NT$1.49 trillion versus NT$1.47 trillion expected, and net income at NT$763.1 billion versus NT$746.34 billion. The 58.7% operating margin also beat the top of TSMC’s own guidance range.
Read the margin line again. More than half of every dollar of revenue landed as net profit. For a business that has to build some of the most expensive factories on Earth, that’s absurd, and it tells you how little pricing pressure TSMC faces at the leading edge.
Where the money comes from
The node mix is the real story. TSMC says 2-nanometer shipments already accounted for 8% of wafer revenue in Q3, with 3nm at 33%, 5nm at 28% and 7nm at 10%. “Advanced technologies” (7nm and below) made up 79% of total wafer revenue.
(A quick decoder: “nanometer” node names are marketing labels for generations of manufacturing technology, not literal measurements. Smaller numbers mean denser, faster, more power-efficient chips, and much more expensive wafers.)
By platform, high-performance computing, the bucket that includes AI accelerators and data center CPUs, made up 68% of revenue, with smartphones at 21%, according to CNBC. Automotive was the fastest-growing segment quarter on quarter, up 23%, though from a much smaller base.
In other words, the overwhelming majority of TSMC’s business is now the stuff that goes into flagship phones, data center processors and AI accelerators. The trailing-edge chips that run your car’s window motors are a rounding error by comparison.
What management said about AI
Chairman and CEO C.C. Wei was about as bullish as a famously measured executive gets. “The AI market continues to be very strong,” he said on the earnings call, pointing to “the rising adoption of Agentic AI as we transition from chatbots to AI assistants” and to greater use of open-source models, which he argued will lower barriers to AI use and drive “more and more demand for leading-edge silicon.”
He also named two newer sources of demand beyond accelerators: CPUs designed for agentic AI workloads, and networking chips. His summary was blunt: “all the AI players are working with TSMC.”
TSMC raised its outlook accordingly. It now expects full-year 2026 revenue growth of about 44% in US dollar terms.
Q4: more
TSMC’s guidance for the fourth quarter is revenue of $52.3 billion to $54.3 billion, with gross margin of 65% to 67% and operating margin of 56.5% to 58.5%, assuming an exchange rate of 31.8 NT dollars to the US dollar.
“Moving into fourth quarter 2026, we expect our business to be supported by continued strong demand for our leading-edge process technologies, including the steep ramp-up of our 2-nanometer technology,” CFO Wendell Huang said in the release.
At the midpoint, that guidance implies roughly 14% sequential growth and a 58% jump from a year earlier, on top of a quarter that already grew 16% in dollar terms. Companies of this size are not supposed to move like this.
The spending: $60 billion-plus and climbing
TSMC reiterated its 2026 capital budget of $60 billion to $64 billion, and said it expects spending from 2026 to 2028 to be “even more significantly higher than the past three years.” Management acknowledged that “tool supply is tight,” meaning the equipment makers TSMC buys from are themselves a bottleneck.
The US footprint keeps growing too. TSMC said construction of its second Arizona fab is complete and equipment is being installed, with volume production now targeted for the third quarter of 2027, slightly ahead of schedule. In July, the company had outlined a total planned investment of $265 billion in Arizona, CNBC noted, and it’s reportedly evaluating Texas as well.
SemiAnalysis analyst Sravan Kundojjala told CNBC that TSMC checks whether customers actually have the financing and supporting components to deploy what they order, and that “even after those checks, TSMC says demand remains well above its supply capacity.”
The catch: the non-AI customer is getting squeezed
The one cautious note on the call is worth more attention than it got. Wei said “consumer and price-sensitive end market segments are being challenged due to the impact of rising component prices and more economic uncertainties,” and that TSMC is “being prudent in our business planning going into 2027.”
Translation: the AI buildout is soaking up capacity and pushing up component prices, and the people buying ordinary phones and laptops are feeling it. TSMC is fine either way, since its premium nodes are sold out. But it’s a hint that the AI boom has costs that show up elsewhere in the electronics economy.
Who wins, who pays
- Wins: TSMC, obviously, and its equipment suppliers, whose tools are on back-order.
- Pays: every chip designer, from Nvidia and AMD to Apple and the hyperscalers building custom accelerators, at prices TSMC largely sets.
- Exposed: consumer-electronics makers caught between rising component costs and price-sensitive buyers.
The takeaway
Our read: every AI lab, hyperscaler and chip designer is fighting over who captures the value of the AI buildout. TSMC is the one company that gets paid regardless of who wins. Nvidia, AMD, Apple, the custom-silicon crowd: they all line up at the same fabs.
The risks are the usual ones, and they haven’t gone away: concentration in Taiwan, export-control politics, tight tool supply, and the possibility that the industry’s spending spree eventually outruns real demand. But if this is a bubble, TSMC is the one selling the soap, at 66% gross margin.
What to watch: whether the 2nm ramp lands inside that Q4 guidance, whether the 2027 capex number climbs again in January, and how long margins this fat can last once the second Arizona fab, with its higher US costs, starts shipping.
Image: 曾 成訓 / Wikimedia Commons, CC BY 2.0
