Taiwan Semiconductor Manufacturing Co. reported third-quarter net profit of NT$763.1 billion, about $24 billion, up 68.7 percent from a year earlier and above the NT$746.3 billion analysts had expected. It was the sixth consecutive record quarter for the world’s largest contract chip maker.
Revenue reached NT$1.49 trillion, about $46 billion, an increase of 51 percent. Operating margin came in at 58.7 percent, above the top end of the company’s own guidance range. The company told investors to expect fourth-quarter revenue of $52.3 billion to $54.3 billion, a midpoint that implies growth of about 14 percent from the third quarter and 58 percent from a year earlier.
The guidance matters more than the quarter. Record results describe demand that has already been booked, while guidance describes what the company believes its customers will order in the next three months. A range that runs well above the prior quarter’s revenue is the company’s answer to the question that has been hanging over the sector for months: whether spending on artificial intelligence computing is still accelerating.
Margins at this level are unusual in manufacturing. A 58.7 percent operating margin means that after the cost of running its fabs, the company keeps close to three of every five dollars of revenue before it accounts for interest and tax. Those margins come from leading-edge production, where the number of suppliers able to build a chip at the required density is effectively one, and where customers pay for certainty of supply as much as for the process.
The company also said it is evaluating an investment in Texas, extending a geographic build-out that already includes Arizona, Japan and Germany. Fabrication outside Taiwan costs more to operate, and TSMC has told customers it expects them to absorb part of that difference. A new site adds capacity that will not arrive for years, which is why customers are also pushing for output from existing plants.
Separately, TSMC said it is willing to cooperate with Terafab, the chip fabrication plant Elon Musk has proposed, while insisting on operational control of its own fabs. That condition is not a negotiating detail. Yield, or the share of chips on a wafer that work, is the asset TSMC sells, and it is produced by tight control of temperature, chemicals, equipment calibration and process recipes. A partner who influences any of those variables can damage the yield, and yield is what customers are buying.
The numbers behind the quarter come from customers spread across the industry. Accelerator designers place the largest single orders, smartphone and personal computer makers place volume orders at older nodes, and the cloud companies that design their own processors have added a bloc of demand that did not exist a decade ago. All of them are competing for the same leading-edge capacity, and the company allocates it in negotiations that customers describe as annual and difficult.
Capital spending is the other side of the record profit. A leading-edge fab costs tens of billions of dollars and takes years to build and equip, so the company commits money before it knows the demand that will fill the plant. Its customers fund part of that investment through advance payments, which ties them to the supplier and gives the supplier visibility that most manufacturers in other industries do not have.
The effect on the wider supply chain follows from those commitments. When TSMC raises its revenue guidance, it signals planned purchases of equipment, wafers, chemicals and advanced packaging capacity, and the suppliers of those inputs read the number as a forecast. The same guidance also tells competitors what they are bidding against, an advantage that comes with being the benchmark.
Beating a consensus estimate by roughly NT$17 billion on profit and carrying an operating margin above the top of its own range suggests the company was conservative when it set that range or that demand firmed after it did. Either reading pushes analysts to raise their numbers for the current quarter, which is what the guidance was meant to do. The risk is that expectations now sit above what the company can deliver in a quarter that is already three weeks old.
Two risks sit inside the results. The first is concentration: a large share of the growth depends on customers whose own revenue comes from a small number of very large buyers of computing. If those buyers slow their orders, the effect reaches TSMC’s fabs within a quarter or two. The second is geography: overseas plants cost more to run, involve more governments and carry timelines that construction schedules rarely honor.
Terafab, if it is built, would be the first fab of scale developed outside the industry’s established makers, and Mr. Musk has described it in terms that imply integration with his other businesses. TSMC’s condition that it keep operating control of its own plants is the same position it has taken with overseas governments that wanted local ownership, and it reflects the fact that the process recipes it has spent decades developing are the product.
The company has not published an outlook beyond the fourth quarter, and its guidance range spans $2 billion, which is wide by its standards. Customers are already asking for capacity years in advance, and the company is deciding how much of its future output to commit before the orders arrive. On Wednesday it said the demand is strong enough to keep building. What it has not said is how long it expects that to last.
