Signal
TSMC Q3 2026: Revenue NT$1,494.2B Up 50.9% YoY, Gross Margin Guided 65-67%
Key Facts
Signal Type
Signal
Industry
AI & Data
Companies
TSMC
Date
October 12, 2026
TSMC delivered its sixth straight record quarter, but the headline number was already known from monthly reports. The real focus is on how the 2nm ramp and overseas fab costs are pressing gross margins, and whether Q4 guidance confirms demand is still accelerating.
Analysts polled by LSEG expect net profit of NT$741-746 billion, up roughly 64-65% YoY. A profit above about NT$707 billion would mark the highest quarterly profit in TSMC's history and an 11th straight quarter of growth.
Q2 wafer ASP reached US$9,271, up 7.8% QoQ. Pricing power is a critical margin support, and any 2027 price commentary could shift forecasts in either direction.
CEO C.C. Wei addressed CoWoS capacity directly on the Q2 call, saying it will be a long time before TSMC can meet customer demand. CoWoS is the advanced packaging step that joins processors and memory into a single AI accelerator, and it remains the industry's primary bottleneck.
CFO Wendell Huang quantified the 2nm hit at about 3 to 4 percentage points of Q3 gross margin. The company has also warned that the full-year dilution could run 2 to 3 points, which is why a margin near the top of the 65-67% range matters more than another revenue record.
Two forces are dragging on margins. The first is the cost of ramping 2nm, where initial yields and volume are still climbing. The second is the higher structural cost of fabs outside Taiwan, which TSMC has consistently flagged as a margin headwind.
3nm lines are running above 100% utilisation, which underscores how tight leading-edge capacity remains even before 2nm contributes meaningfully to revenue.
The capex increase to US$60-64B for 2026, and Morgan Stanley's projection of US$75-80B for 2027, can be read two ways. Some investors see heavier spending as proof that demand is real and accelerating. Others see it as a cash drain that weighs on free cash flow before the revenue materialises.
TSMC's pricing power helps bridge that gap. A 7.8% QoQ rise in wafer ASP to US$9,271 shows the company can pass higher costs to customers. Any commentary on 2027 pricing will be closely parsed for whether that leverage continues.
The potential second US hub near Dallas, with up to six advanced fabs, could cost more than the US$265 billion already committed to Arizona. TSMC has not confirmed the plan, but analysts expect questions on the call. A larger US footprint would deepen exposure to higher overseas costs while addressing tariff and geopolitical pressure.
The Section 232 tariff of 25% on certain advanced AI chips took effect on 15 January 2026, covering chips such as Nvidia's H200 and AMD's MI325X. Wide exemptions apply, including chips for US data centres, but the tariff adds a policy layer to every capacity decision TSMC makes outside Taiwan.
Terafab, the Texas-based project bringing together Tesla, SpaceX, and xAI with Intel as the first manufacturing partner, adds another dimension. Elon Musk confirmed early talks with TSMC on 3 October. In the near term, Terafab could become a large customer. Over a longer horizon, it could also create a new rival if the project scales its own manufacturing capability.
The OpenAI revenue report on 8 October knocked the Nasdaq down 1.25% and chip stocks down 3.4%. OpenAI's annualised revenue was close to US$50 billion at the end of September, about US$20 billion below earlier circulated figures. Part of the gap is accounting, since OpenAI does not count sales through cloud partners as revenue. Even so, the news revived concerns about circular AI spending between model makers and chipmakers.
UBS takes the other side, citing industry data that still points to robust AI compute demand. The bank expects average hyperscaler cloud revenue growth of 58% in the September quarter, up from 48% in June, while warning that growth and financing worries could cause volatility.
Samsung's preliminary Q3 numbers offer a cautionary parallel. Operating profit jumped almost ninefold, yet shares fell because analysts had trimmed forecasts as the won strengthened. For TSMC, that episode shows how a record can still disappoint when expectations are this elevated.
I would watch four things over the next two quarters:
Source:
AdmiralsGet AI signals in your CRM
Regulation, model launches, funding mega-rounds, acquisitions, privacy lawsuits, and compute constraints.
