Record breaking financial report, unprecedented expansion of production: The global production capacity game behind TSMC's third quarter report

On October 15th, TSMC officially released its financial report for the third quarter of 2026: consolidated revenue of NT $1494.24 billion, net profit after tax of NT $763.06 billion, and earnings per share of NT $29.43 billion. These three indicators simultaneously broke historical records, marking the sixth consecutive quarter in which the company has rewritten its quarterly profit peak. In US dollars, the third quarter revenue was $46.72 billion, a year-on-year increase of 41.2% and a month on month increase of 16.2%, significantly exceeding the previous financial forecast range of $44.6 billion to $45.8 billion. As the global debate over the sustainability of capital expenditures on artificial intelligence continues to heat up, this report from the top of the industry chain provides a rare "hard data" validation for the entire narrative of AI computing power.

This unexpected performance reflects that the demand for AI computing power has been effectively transmitted to the manufacturing end. The proportion of advanced processes continued to rise this quarter: 2nm processes contributed 8% of wafer sales revenue for the first time, while 3nm, 5nm, and 7nm accounted for 33%, 28%, and 10% respectively. The total proportion of the four types of processes reached 79%, and advanced processes have become TSMC's absolute revenue driver; Corresponding to the downstream application sector, high-performance computing has become the largest single application sector with a proportion of 68%, a month on month growth of 22%, smartphones account for 21%, and the remaining sectors total about 11%. Among them, consumer electronics only accounts for 1% and has declined by 12% month on month, indicating significant differentiation in downstream demand. The high prosperity of AI computing power contrasts sharply with the weakness of the traditional consumer market.

The explosive growth of AI business has further highlighted TSMC's crucial position in the entire industry chain. Whether it's Nvidia's general-purpose GPU, Broadcom's customized ASIC, or various cloud service providers' self-developed acceleration chips, their physical carriers must ultimately be on advanced process wafers. Currently, there is only one global manufacturer capable of stable and large-scale production of 3nm and below processes. In other words, TSMC's wafer production capacity has become the "main gateway" for global AI computing power supply. When the model parameters and computing power cluster continue to expand, downstream demand will be transmitted layer by layer along this pipeline, and a large number of orders will be converted into TSMC wafer orders, directly reflected in the capacity utilization rate.

However, there are also hidden concerns under the strong demand. The most obvious manifestation is that the gross profit margin in the third quarter decreased by 1.3 percentage points month on month, ending the previous continuous upward trend. The factors that cause profit pressure mainly include two layers: one part comes from the dilution of gross profit in the early stage of 2nm process mass production - the new process yield and capacity utilization rate have not yet climbed to the point, but depreciation has been fully included in costs, and it is necessary to exchange profits for capacity first; The other part comes from the dilution effect of overseas wafer fabs. The management of TSMC stated that overseas production will drag down the consolidated gross profit margin by nearly 2 percentage points in 2026. As more overseas factories are put into operation, this drag may expand to 3 to 4 percentage points in the future. The simultaneous increase in equipment, material, and labor costs is also continuing to push up the depreciation burden. The company's response to this is very clear: on the one hand, it can hedge by adjusting prices, improving capacity utilization and manufacturing efficiency, and on the other hand, it can adhere to the operating bottom line of "revenue growth rate higher than capital expenditure growth rate".

Behind the short-term profit pressure is TSMC's unprecedented global production expansion plan, which is also the core issue that this financial report has more affected the market's nerves. TSMC used to manage 4 to 5 factory building projects at the same time. Today, nearly 20 wafer fabs around the world are advancing at the same time -13 advanced manufacturing and packaging plants in Taiwan, China, China, and 5 to 6 overseas. Correspondingly, there is a continuous increase in capital expenditures: expenditures will remain in the range of $60 billion to $64 billion in 2026, and management has extended visibility to 2028, clarifying that the total expenditures for the three years 2026-2028 will significantly exceed the total of approximately $101 billion in 2023-2025. What supports this investment intensity is the management's firm judgment on the demand outlook - TSMC Chairman Wei Zhe jia bluntly stated at the press conference, "We do not intend to leave business opportunities to others

It is worth mentioning that the significance of this round of expansion goes far beyond simply increasing production capacity. It is also a strategic layout that deeply binds the three things of diversifying geopolitical risks, obtaining government subsidies, and locking in long-term customer orders. In the past, TSMC was a highly concentrated wafer foundry in a single region, but now it is transforming into a distributed manufacturing network with "global layout and core retention in Taiwan": by establishing production clusters in the United States, Japan, and Germany, it not only resolves customers' demands for supply chain security, but also directly exchanges policy resources from various countries, and keeps competitors outside the dual barriers of technology and production capacity. The cost is the decline in capital efficiency - overseas factories have completely different cost structures in labor, construction, infrastructure, and local supply chains, and economies of scale and yield ramp up require time.

In fact, in addition to the internal cost pressure brought by expansion itself, there are two variables that will determine the future direction of expansion. Firstly, for high-end AI chips, advanced packaging capacity is becoming another key constraint after wafer manufacturing. Who can obtain CoWoS production capacity largely determines who can obtain the shipment volume of AI chips, and the focus of industry competition is extending from process to packaging, substrate, and testing. TSMC has currently achieved mass production of 5x mask size packaging, with excellent yield performance, and is advancing towards 9x and 14x; Secondly, there is uncertainty in terminal demand. Wei Zhejia admitted that the consumer and price sensitive terminal market is facing pressure from rising component prices and macro uncertainty, so the company remains cautious when planning its business for 2027. This means that in the future, there may be two completely different situations: if the demand for AI continues to increase, the capacity gap will support TSMC to maintain high gross profit and pricing power; If terminal demand is suppressed by rising costs, the depreciation pressure of new production capacity will quickly become apparent, and the growth logic of AI capital expenditure will also face a real stress test.

When the artificial intelligence competition enters the deep water zone, advanced process capacity has become the most scarce "hard currency" in this competition. TSMC's third quarter financial report is not only a record breaking report card, but also a strategic roadmap for global production capacity layout. In the next two years, the real focus will not be on TSMC's short-term profits, but on whether it can convert its investment in nearly 20 factories into sustained leading technology and stable returns - and those players who have locked in advanced processes and packaging capacity in advance will also have an advantage in the next wave of artificial intelligence infrastructure.