According to multiple people familiar with the matter, TSMC is asking customers to plan and apply for its N2 process capacity quota as early as possible, as far as the second quarter of 2027, and most of the production capacity in the next two years is close to being sold out. Although some production capacity is still available for reservation, industry insiders say that the quota lock-in period is being lengthened to as many as six quarters, which makes chip design companies face increasing uncertainty when formulating their own product roadmaps.
A number of investment bank analysts, including Deutsche Bank and JPMorgan Chase, have previously pointed out that TSMC’s N3 process capacity has been booked by customers beyond 2027. Many media quoted industry sources as saying that the 3-nanometer production line has been at "full capacity" for a long time. Industry news this time shows that the N2 process, which has just entered mass production, is quickly heading towards a similar tight capacity situation and will continue into next year and even beyond.
TSMC will officially enter mass production of the N3 process in 2022. This node technology almost doubled the company’s revenue last year to about $25 billion. In contrast, the N2 process only started mass production in the past six months, and Apple is believed to be one of the first customers and is expected to be the first to use this process on upcoming MacBook chips. TSMC has also planned an upgraded version of N2, N2P, which will provide about 5% performance improvement under the same power consumption. The mass production time is scheduled for the second half of 2026; at the same time, the A16 process for high-performance computing chips will also enter the mass production stage during the same period.
For a long time, competing for TSMC's advanced process production capacity has been a "required course" for global chip design companies. This issue was particularly highlighted during the COVID-19 epidemic, when car companies were forced to reduce production due to insufficient supply. However, many industry insiders said that the current tightness of production capacity is more severe than in the past. On the one hand, high-end chips such as AI GPUs have a larger single wafer area and occupy more production line resources; on the other hand, the unit price of such chips is extremely high, and customers are increasingly accepting of TSMC's rising foundry prices. This, coupled with the "arms race" between AI startups and ultra-large-scale cloud service providers, has jointly increased the intensity of the competition for advanced process production capacity.
Previous analysis pointed out that in terms of changes in the number of TSMC's largest customers, Nvidia has surpassed Apple, which it has cooperated with for more than 15 years, in several quarters of 2025, becoming the company's largest customer in the world for some periods of time. The latest data further shows that Nvidia's purchases from TSMC more than doubled year-on-year in 2025, reaching approximately US$23.3 billion, surpassing Apple's US$20.7 billion.
As usual, TSMC will require customers to declare the number of wafers required on a quarterly basis, and then plan capacity allocation and confirm specific quotas for each customer based on this. When customers finally lock in production capacity, they also need to pay corresponding fees to ensure production schedules. The current lock-in "deadline" is about 6 months ahead of the start of construction, and wafer manufacturing and back-end packaging usually take 4 to 6 months, which means that chip design companies often have to decide their production capacity needs about 12 months before they are expected to get finished products.
Although TSMC has also retained the so-called "hot lots" mechanism for some customers, which means jumping in line to arrange emergency production at a high premium, the number of such orders is limited and not guaranteed. As overall production capacity continues to tighten and delivery cycles continue to lengthen, customers in downstream computing, networking, consumer electronics and other fields are increasingly concerned about whether mid- to long-term supply can be implemented as planned.
This week, Nvidia provided side evidence of its future capacity needs when it released its financial report: the company's full-year revenue in 2025 increased by 65% year-on-year to $216 billion. Colette Kress, Nvidia's chief financial officer, said at an investor call that the company has covered shipment needs extending to 2027 through inventory and supply commitments. She also pointed out that although it has not yet used the expression "all production capacity will be sold out in 2027," the current visibility for the next few years is "unprecedentedly clear" because customers know that if they do not lock in capacity now, they will be left behind in the "agent AI" race.
In order to alleviate the increasing pressure on production capacity, TSMC recently announced that it will add an N3 process production line at its Kumamoto factory in Japan to further disperse and expand the advanced process production capacity layout. Previously, TSMC has also added a third phase expansion plan for its factory in Arizona, USA. It will produce N2 and A16 process wafers locally in the future. The new factory is expected to achieve mass production by the end of this decade. At the same time, the company continues to promote the construction of new factories in Taiwan to cope with the growing long-term global demand for advanced manufacturing processes.