TSMC's Record $23.6B Profit, Judged Against Perfection

By the TechPulse USA News Desk

Taiwan Semiconductor Manufacturing Company made more money in a single quarter than most companies make in a decade — and the stock was judged against perfection. On Thursday, the world's most important chipmaker reported a third-quarter net profit of NT$763.1 billion (US$23.6 billion), its sixth consecutive record quarter, up 68.7% from a year ago and comfortably ahead of the LSEG SmartEstimate of NT$746.3 billion. The American depositary shares had already closed Wednesday down 3.4% at US$430. By the time the numbers hit the wires in Taipei, the stock had already voted.

This is what "judged against perfection" looks like in practice. TSMC shares have ballooned nearly 60% this year, the company sits at a market capitalisation of roughly US$2 trillion — nearly double Samsung's — and options pricing implied a move of roughly 4% through the end of the week even before Thursday's release. When you are priced for immortality, merely beating estimates by US$500 million does not excite anyone. We cross-checked Reuters' figures against the Business Times and Invezz's same-day reporting while writing this piece; every number matched to the decimal. The numbers are not the argument here. The argument is what the market demands next.

Photo: TSMC Fab 18 in the Southern Taiwan Science Park — one of the advanced-node plants feeding the AI chip boom. Photo: 4300streetcar via Wikimedia Commons (Creative Commons, attribution).

The scoreboard: revenue was the sideshow

Start with the arithmetic, because it is genuinely absurd. Revenue for the July-to-September quarter reached NT$1.49 trillion, up 50.9% year on year — the revenue record we covered when it was announced on October 9. In US dollars that was US$46.72 billion, up 41.1% from a year earlier and a touch above the US$45.2 billion mid-range guidance figure. Quarter on quarter, revenue grew 17.6% while net income grew 8.0%, per the company's consolidated accounts. Diluted earnings per share came to NT$29.43, or US$4.60 per ADR unit — a beat of nine cents on that measure.

The revenue number mattered less than it usually would, for a reason worth stating plainly: it was already known. TSMC publishes monthly sales figures, so September's NT$511.86 billion tally (up 54.6% year on year) had already let analysts assemble the quarter's total before Thursday. The real event was always going to be profitability — whether the machine that prints money at this scale could keep printing it while building the most expensive factories on Earth.

It could. Gross margin came in at 66.4%, squarely inside the 65–67% guidance range. Operating margin hit 58.7%, above the top of management's 56–58% forecast. Net margin stood at 51.1%. For context, that means that of every dollar TSMC collects, more than fifty cents flows through to net profit — a figure that would be embarrassing for a software company and is frankly startling for a manufacturer that melts silicon for a living. Bank of America analyst Haas Liu had expected gross margin near 67% ahead of the report and described TSMC's valuation at about 16 times estimated 2027 earnings as "undemanding," according to Invezz's account of TipRanks data. Undemanding. Sixteen times forward earnings, for the monopoly at the centre of the AI century.

The margin story: the 2nm ramp didn't bite

Margins were the quarter's real plot, because the bear case going in was entirely about them. The 2-nanometre process is in steep ramp — its share of wafer revenue jumped to 8% from just 3% in the previous quarter, per TrendForce's read of the release — and new nodes are supposed to be margin-dilutive while fabs absorb costs ahead of maturity. Add the higher operating costs of overseas plants, and several previews had sketched a quarter in which profitability would finally bend.

It didn't bend. The mix carried it: 3-nanometre chips contributed 33% of wafer revenue, 5-nanometre another 28%, and 7-nanometre 10%, meaning 79% of wafer sales came from advanced technologies. That is the answer to the dilution question — when your customers are paying premium pricing for the world's most advanced nodes, and when even the "mature" nodes you sell are mostly leading-edge by anyone else's standards, ramp costs get absorbed by pricing power. Operating profit rose 75% year on year to NT$876.4 billion, with the operating margin climbing from 50.6% a year earlier to 58.7%. ChosunBiz's write-up of the results noted the 8.1-percentage-point year-on-year expansion as the headline of the quarter. It is worth saying directly: an 8-point margin expansion at NT$1.49 trillion of revenue is colossal. It is the kind of number that would be a whole turnaround story at another company; at TSMC it is just Thursday.

Photo: TSMC Fab 14B in Tainan, in the Southern Taiwan Science Park, where advanced-process capacity continues to expand. Photo: 4300streetcar via Wikimedia Commons (Creative Commons, attribution).

The perfection problem

So why the shrug? Part of it is Thursday's tape, not Thursday's numbers. Taipei-listed shares closed down 2% ahead of the release with the benchmark off 3.4%, and the US-listed ADR had already fallen 3.4% on Wednesday as part of a broader technology selloff — hot US inflation data revived worries about further Federal Reserve rate hikes, and Fed funds futures priced the odds of a quarter-point hike at the October 28 meeting near one-in-three. AI-linked names bleed first when rates expectations turn hawkish, because the trade runs on borrowed money and borrowed time.

The deeper part is the AFP wire's phrase that the Straits Times carried: chipmakers reporting bumper results risk being "judged against perfection" by investors. Perfection pricing is the defining pathology of the AI trade in 2026. The same week, Anthropic reportedly turned a profit for the first time in its history — and Wall Street sold it off anyway, because adjusted profitability next to hundreds of billions in compute commitments looked like cosmetics rather than proof. TSMC is now getting the mirror-image treatment: real, audited, colossal profitability — and a market that treats it as table stakes. The irony is that TSMC's profit is the thing that proves the AI economy actually generates cash. Nobody is cheering that yet.

Zavier Wong, market analyst at eToro, told AFP ahead of the announcement that TSMC "sits at the very centre of the AI ecosystem," building chips for Nvidia, Apple, AMD and Broadcom. He is right, and the centre is a strange place to be this week: everything flows through you, nothing is credited to you, and your reward for the best quarter in your company's history is a 3.4% decline and a market asking what you intend to do for an encore.

What the earnings call has to answer

The initial release contained no fourth-quarter revenue or margin guidance — that arrives on the earnings call, scheduled for 0600 GMT on Thursday — and the call is now the whole ballgame. Two questions will decide whether the record profit is the start of a rerating or the peak of a cycle.

First, full-year growth. Morgan Stanley analyst Charlie Chan expected management to raise its 2026 revenue-growth outlook into the low-to-mid 40% range from guidance of more than 40%, according to Reuters' preview. TSMC's own September disclosure had put January-to-September revenue growth at 41.1% year on year, so the arithmetic is already pulling upward; the question is whether management confirms it publicly. A raised guide would signal that demand from Nvidia, Apple and the custom-silicon houses is strong enough to absorb the cost of the new capacity wave.

Second, capital spending — the subject that never leaves the room. TSMC raised its 2026 capex forecast in July to US$60–64 billion, from a previous US$52–56 billion, as it expands advanced nodes, packaging capacity and overseas fabs. That is the industry's other supply-chain story: ASML told the world this week it cannot build lithography machines fast enough to meet demand through 2028, and TSMC is the customer standing at the front of that queue. Spending discipline matters because the macro backdrop has turned squeamish — Wednesday's hot CPI print is a reminder that the cost of capital is rising again, and rising capital costs are the one thing that can break an expansion plan this large.

The American question: Arizona, and maybe Texas

Hovering over every guidance number is the American build-out, and it is enormous enough to deserve its own section. TSMC has committed US$265 billion to its Arizona manufacturing complex — including the additional US$100 billion unveiled in July — and Reuters reported in September that the company is evaluating a potential investment in Texas, a story we covered in depth after Taiwan's Economic Daily News put the Dallas area under discussion.

Texas remains evaluation, not commitment, and the analysts are notably unhurried about it. Morningstar analyst Phelix Lee told Reuters that Texas was "a good location for a 2030s or 2040s expansion grand plan," while seeing limited urgency given Arizona capacity is still being built. That framing matters: the American question for Q4 is not whether TSMC will announce a Texas megaproject on the call — it won't — but whether management shows that Arizona's cost curve is bending toward respectability as volume ramps. The overseas-fab margin dilution everyone fears is the one number in this report that investors cannot verify from outside. Thursday's call is where it gets asked, in so many words.

Our take: the market's shrug is a confession

Let's call this quarter what it is. TSMC has now posted eleven consecutive quarters of profit growth and six straight records. It prints a 51% net margin while ramping the most advanced manufacturing process in human history, in a year when its American customers are re-shoring production, its European equipment supplier is rationing machines, and the Federal Reserve is making borrowing more expensive by the month. The company has not botched a single thing. The quarter is, by every honest measure, a triumph.

And the market's response — down 3.4% before the numbers even printed — is a confession about who is really under strain. It is not TSMC. It is the narrative that justifies today's valuations: the belief that AI spending compounds forever, that the cost of capital stays cheap, that every record quarter will be followed by a bigger one. TSMC cannot control any of those things. It can only do the thing it has done, again, better than anyone on Earth does anything: build the chips.

The shoddy part of this week's trading is not the selling; markets sell for macro reasons every day. The shoddy part is the pretence that TSMC must answer for it. There is a company at the centre of the AI ecosystem that just reported the largest quarterly profit in its history, beat estimates, expanded margins, and is already building the factories that decide where the 2030s get their compute. If investors cannot find anything to like in that, the problem is not the quarter. It is the expectations — and expectations, unlike wafers, do not get better with yield.

Sources

Reporting: Reuters on TSMC's Q3 profit and October 14/15 earnings previews (reporting by Wen-Yee Lee and Ben Blanchard, editing by Thomas Derpinghaus); the Straits Times' AFP-sourced account of the results (October 15, 2026); the Business Times' same-day write-up on the sixth straight record quarter (October 15, 2026); ChosunBiz's Korean-language results coverage with margin detail (October 15, 2026); Invezz's earnings analysis by Devesh Kumar (October 15, 2026); TrendForce's results summary with node-share breakdown (October 15, 2026); Seeking Alpha's ADR earnings summary (October 15, 2026). TSMC's own figures are on a TIFRS consolidated basis for the quarter ended 30 September 2026. Analyst quotes as attributed to their outlets; Texas evaluation per Reuters' October 14 preview.

By TechPulse USA — TechPulse USA is an independent tech news publication covering the global AI race with a US-first lens. We verify every fact we print, link our sources, and separate what is proven from what is claimed.

Update, Oct. 15: the earnings call delivered

TSMC's Q3 earnings call — held this morning, Oct 15 — answered both open questions this post flagged. Q4 2026 revenue guidance: US$52.3–54.3 billion — 14% sequential growth at the midpoint, and a roughly 58% year-on-year leap at the midpoint. Q4 margin guidance: gross margin 65–67%, operating margin 56.5–58.5% — the 2nm ramp dilution management warned about is in the numbers, not the surprise. Full-year 2026 revenue growth guidance raised to about 44% in US-dollar terms (up from "more than 40%"). 2026 capex reiterated at US$60–64 billion; spending from 2026 to 2028 will be significantly higher than the prior three years, with nearly 20 fabs being built worldwide, per call coverage.

Chairman C.C. Wei on AI demand: the gap between demand and supply is, in his phrase, "still quite big" — the demand story described above is intact. New wrinkle: overseas-fab margin dilution will widen to 3–4% in coming years (from 2%) as US and global expansion accelerates. And on Elon Musk's Terafab: Wei said TSMC is in discussions with SpaceX's Texas chip-factory project, but that any fab TSMC uses "has to be run by TSMC" — Musk had written Oct 7 that TSMC could sublease part of Terafab "but nothing more." HPC was about 68% of revenue — the AI infrastructure concentration keeps rising. The market wanted perfection; it got a guidance raise and a demand-supply gap that is still, in the chairman's word, big. (All call figures as reported by call coverage this run.)