Taiwan Semiconductor Manufacturing reported another set of record numbers on October 15, the sixth consecutive quarter in which it has set a high for revenue and profit, and raised its expectations for the year ahead.
Revenue reached 1.4942 trillion New Taiwan dollars, up 50.9 percent from a year earlier. Net income was 763.06 billion New Taiwan dollars, an increase of 68.7 percent. Earnings per share came to 29.43 New Taiwan dollars, or 4.60 dollars per American depositary receipt, both above the average of analyst estimates.
The margin is the figure that separates this cycle from previous ones. Operating margin reached 58.7 percent, at the top of the range the company had guided to. A foundry that earns that much on each dollar of sales is pricing scarce capacity, and scarcity has been the defining condition of the market for accelerators and the memory that feeds them.
Guidance for the fourth quarter pointed at revenue of 52.3 billion to 54.3 billion dollars, a midpoint roughly 58 percent above the year-earlier period. For the full year the company lifted its expected growth in dollar revenue to about 44 percent from slightly above 40 percent, and kept capital spending in a range of 60 billion to 64 billion dollars for 2026.
Capital spending at that level is the clearest statement about what the company expects demand to do. A foundry builds capacity years before it knows the price it will charge, and the decision to keep spending at this scale in a quarter when semiconductor shares fell sharply reflects contracts and customer commitments rather than sentiment about the sector’s valuation.
The technology mix is moving faster than the company’s own plans assumed. Two-nanometer production accounted for 8 percent of wafer revenue in the quarter, up from 3 percent three months earlier. Each new node carries a higher price per wafer and a period during which only one manufacturer can supply it, which is where the margin comes from.
Management described demand from inference and from agents as outstripping its ability to supply central processors. That is a change in the composition of demand. Training large models has been the driver behind the past several years of accelerator purchases. If deployed systems that answer queries consume processors in volume, the demand base widens and becomes less dependent on a small number of very large buyers.
The supply response is constrained in ways that money does not fix quickly. Advanced fabrication requires lithography equipment with multi-year lead times, a workforce trained to run the tools, and power and water at the site. Each of those is a separate bottleneck, and a company that has already committed its capacity forward is limited by the slowest of them.
Customers have responded by trying to secure capacity in advance rather than by shopping on price, and the company’s own executives have described the volume of inquiries in those terms. When buyers are asking for allocation instead of quotes, the pricing conversation changes: contracts lengthen, prepayments appear and the supplier’s negotiating position strengthens for the duration of the build.
The results also say something about geography. Most of the world’s most advanced logic is produced in Taiwan, and every government that wants artificial intelligence capacity inside its borders has to deal with that concentration. The company’s expansion plans abroad, which are slower and more expensive per wafer than its home plants, are being priced into its capital budget.
Analysts had expected a strong quarter and still had to raise their numbers. The pattern of the past six quarters is that guidance issued three months earlier turns out to have been conservative, which is itself a signal about how the company reads its own order book.
What the report does not settle is how long the current level of demand lasts. Memory prices have risen sharply, power constraints are binding in several markets and the buyers of accelerators are spending against revenue forecasts rather than against realized demand. Foundries will build what customers commit to, and the commitments that matter for the next two years have already been made.
Currency moves add a variable the company does not control. Revenue reported in New Taiwan dollars depends on the exchange rate as well as on volumes and prices, and a changing rate shifts the reported growth without changing the number of wafers shipped. The sensitivity is why the company guides in dollars and discusses exchange assumptions alongside its revenue ranges on the same call.
The company also sells to customers whose own forecasts are shifting. Designers of custom accelerators, makers of networking gear and the companies building power infrastructure around new data halls all place orders that depend on projects they have not finished funding. A foundry sees those orders before the projects are complete, which gives it an early read on demand that its customers lack.
