TSMC's sixth straight record quarter — 58.7% margin, 68% from HPC

Share
TSMC's sixth straight record quarter — 58.7% margin, 68% from HPC

The chip cycle's upstream end and its downstream end told two different stories on the same morning: TSMC booked a sixth consecutive record, while the lab that buys its wafers told investors its growth rate is falling fast.

TSMC's third-quarter net profit rose 68.7% to NT$763.1 billion (about $23.9 billion), a sixth consecutive record, on revenue of NT$1.494 trillion — up 50.9% year over year and roughly $46.7 billion. Both lines beat the LSEG SmartEstimate consensus (NT$1.46 trillion revenue, NT$740.8 billion net income), and the operating margin came in at 58.7%, above the top of the company's own 56–58% guidance — which is the clearest signal of pricing power in the whole report. The mix says the same thing. High-performance computing was 68% of platform revenue against 21% for smartphones, meaning the AI buildout is now the base of TSMC's business rather than a growth segment attached to it. Three-nanometer wafers were 33% of wafer revenue, and automotive was the fastest-growing platform at 23% quarter over quarter. Free cash flow jumped to NT$476.6 billion from NT$139.4 billion a year earlier. The company has committed $265 billion to Arizona and is reportedly weighing a Texas site too.

The reason to read this as a lagging indicator rather than a forecast: TSMC books revenue when wafers ship, which is six to nine months after the order that paid for the capacity. So a record quarter tells you what the AI trade looked like in the spring. What got repriced on Wednesday was the order book ahead of it. Anthropic told pre-IPO investors its third-quarter revenue reached $16.8 billion — up about 46% from $11.5 billion in the second quarter, but a visible deceleration from the more than 140% sequential growth it posted in Q2, and the company said it was profitable on an adjusted basis. MarketWatch framed the slowdown as the AI trade's latest red flag and semiconductor shares slid after the numbers circulated.

Two days ago the market sold off memory and server names on a Barclays downgrade; today the industry's most leveraged supplier posted its best quarter ever and the semis still sagged. That is what a cycle top looks like from the inside — not a bad quarter, but a good one nobody wants to pay up for. The one thing that would change the read is TSMC's own guidance for the current quarter and its capex number for 2027, neither of which lands with these results.


AI agents made more than 138 million payments on Base through Coinbase's x402 stablecoin protocol — and the median new wallet now spends 18 minutes after it is funded. The numbers come from Chainalysis's 2026 Geography of Crypto report: over 87% of x402 payments were worth between a hundredth of a cent and a dime, the share of wallets making their first payment within an hour of funding climbed from 4% in the third quarter of 2025 to 67% a year later, and repeat funding within 30 days rose from 70.6% to 89.3%. Geographically, US-linked wallets accounted for 88.1% of identified payments for AI model inference, against 0.1% for China-linked ones. Fortune put the cumulative transaction count at roughly 1,500 in mid-2025, when the protocol was weeks old. Chainalysis itself calls agentic commerce an "experimentation phase."

The catch worth carrying: transaction count is the one metric an interested party can manufacture almost for free, since the payment facilitator sponsors the gas and nothing on-chain records who actually controls a wallet. An arXiv paper measuring the same chain over 280 days found 136.7 million settlements worth $44.1 million in total — about $0.32 each — of which 21.2% were fictitious and 63.8% were internal settlement inside a linked cluster. Genuinely independent value, by the authors' accounting, lies between $187,861 and $20.3 million. Their conclusion is blunt: "Settlement count measures manufacturability, not adoption." Both things are true at once — the rails work, at a price point card networks cannot reach, and almost nobody outside a few circles is using them yet. We wrote about agentic commerce's first legs in September, when Alipay's payment robot dog went shopping, and the pattern held from the traditional-payments side: real machinery, thin volume.

What to watch: whether USDC volume on x402 keeps growing once the incentive-subsidized wallets roll off, and how TSMC guides capex for 2027 — the memory fab buildout it depends on is where the deceleration would show up next.

Would you trust an agent with a funded wallet and no spending limit — and if not, what limit would you actually set? Write to us via our contact page.

Get SiliconNoon in your inbox

One email a day with the AI stories that matter — what happened, the numbers, and why it matters. No spam, unsubscribe in one click.

Read more

Huawei's physical-AI report grades cross-layer compute at 2 out of 5

Huawei's physical-AI report grades cross-layer compute at 2 out of 5

Huawei wants to sell the industry a coordination layer for physical AI. Its own report puts that layer in the least mature row of the table. Huawei's Physical Intelligence Cloud-Edge-Device Collaboration Architecture report, released on 18 September at HUAWEI CONNECT 2026 in Shanghai, scores cross-layer heterogeneous compute scheduling — the thing the whole architecture is for — at 2 on a five-level maturity scale, the lowest grade in the report's own technology table.

Salience Labs raises $125M for all-optical AI data center switches

Salience Labs raises $125M for all-optical AI data center switches

The AI buildout's bottleneck moved from the processor to the wire, and today's money followed it: a British photonics startup pulled in a nine-figure round with both Intel and AMD on the cap table, while two smaller bets went to the less glamorous end of the agent economy. Salience Labs raised an oversubscribed $125 million Series B led by Bullhound Capital, with Intel Capital and AMD Ventures both participating — two chipmakers that compete for the same accelerator sockets, funding the same…

NTT Data in talks to build Anthropic's first Japan data center

NTT Data in talks to build Anthropic's first Japan data center

The overnight inbox is an infrastructure story dressed as a rumor, plus two research-side reads: a paper arguing capability audits have a cryptographic hole, and Ben Recht taking a hatchet to OpenAI's math dump. NTT Data is in advanced talks to build a data center for Anthropic in Osaka with capacity of up to 500 megawatts — among the largest in Japan, with construction costs potentially reaching 1 trillion yen ($6.3 billion).

SpaceX reportedly seeks $40B, led by Apollo, to buy Nvidia chips

SpaceX reportedly seeks $40B, led by Apollo, to buy Nvidia chips

The compute boom found its lender, safety regulation found its loudest advocate, and the open-source world got a model that only picks answers. SpaceX is reportedly lining up $40 billion, led by Apollo, to buy Nvidia chips — and almost all of it is debt. The Financial Times broke the plan, citing people familiar with it; Reuters and Bloomberg have since relayed the report, and SpaceX, Apollo, Nvidia and Pimco have all declined to comment, so this is reported rather than confirmed.