DigiTimes复盘台积电Q3法说会：Q4营收指引季增约15%、全年美元营收+44%，并重申评估马斯克Terafab但无任何协议

TSMC chairman C.C. Wei. Credit: DIGITIMES

TSMC reported record earnings for the third quarter of 2026 and forecast a further 15% increase in fourth-quarter revenue, putting full-year US dollar revenue growth on course to reach 44%. Yet supply chain sources say the earnings call carried a more consequential message: TSMC's increasingly explicit assertion of the bargaining power afforded by its lead in advanced process technology and manufacturing scale.

TSMC has confirmed plans for N3 fabs in Taiwan, the US, and Japan, while capacity for its N2 process family is expected to more than double in 2027 from 2026. Expanding advanced-node production serves two purposes: meeting demand for AI, high-performance computing (HPC), and high-end chips, while reducing incentives for customers to turn to rival foundries when capacity is constrained.
Industry sources say TSMC's influence now extends beyond manufacturing to order allocation, wafer-start scheduling, and pricing negotiations.
Four new N3 fabs strengthen TSMC's supply and customer retention
TSMC expects full-year capital expenditure to remain at US$60–64 billion. Fourth-quarter revenue is projected at US$52.3–54.3 billion, with the midpoint representing approximately 14% sequential growth and 58% year-over-year growth, underscoring management's confidence.
Chairman C.C. Wei emphasized that TSMC's current strategy is to produce as much as possible, as fast as possible. The company aims to accommodate as much customer demand as possible rather than leave available business opportunities to competitors, reducing the risk of customers switching foundries while awaiting capacity.
TSMC began mass production of its N2 process in the fourth quarter of 2025 and continues to expand capacity in Hsinchu and Kaohsiung. Production capacity for the N2 family is expected to more than double in 2027 from 2026. Its next-generation A14 process is scheduled for risk production in 2027, followed by mass production in 2028.
For N3, TSMC is expanding its manufacturing footprint across Taiwan, the US, and Japan, with additional capacity planned in Tainan, Arizona, and Japan. An N3 fab already under construction in Tainan is scheduled to begin mass production in the first half of 2027, while another newly announced fab is targeted for the first half of 2028. In Arizona, construction of the second fab has been completed, with equipment move-in and installation underway ahead of mass production in the third quarter of 2027. Japan's second fab will adopt N3 technology, with production expected in 2028.
Beyond new facilities, TSMC plans to convert some existing 5nm equipment in Taiwan to support N3 production and optimize capacity allocation across its 7nm, 5nm, and 3nm nodes. New fabs will provide medium- to long-term capacity, while equipment adjustments offer near-term flexibility and improve utilization of existing assets.
Supply chain analysts say insufficient capacity is often a key reason customers consider alternative foundries. By addressing N3 supply constraints, TSMC could enable customers to increase wafer orders within their existing process and supply chain arrangements, making it harder for competitors such as Samsung Electronics and Intel to capture established orders solely by offering available capacity.
However, new fabs require time to achieve stable production. Whether TSMC's expansion can meaningfully discourage customer migration will depend on when additional capacity becomes available and how demand evolves.
N2 capacity to more than double in 2027, raising the cost of switching foundries
For GPUs, AI accelerators, and custom chips developed by major cloud service providers, changing foundries involves considerably more than replacing a supplier. Product redesign, photomask preparation, process qualification, yield improvements, and subsequent mass production all take time. Even when another foundry is willing to accept an order, replicating existing manufacturing conditions is not necessarily immediate.
Supply chain sources say foundries allocate capacity according to customers' product development schedules, wafer-start plans, and order commitments. If customers delay wafer starts, reduce demand, or fail to confirm orders on schedule, previously reserved capacity may be reassigned to other customers.
Customers that switch suppliers and subsequently seek to increase orders with TSMC may therefore need to renegotiate capacity allocations and production schedules. Once their original slots have been allocated elsewhere, additional wafer orders cannot necessarily resume under the previous timetable. This is what supply chain sources describe as having to "rejoin the queue."
Persistent constraints on advanced-node capacity have also increased the importance of early reservations, prepayments, and capacity commitments. However, prepayments do not entitle customers to immediate production capacity on demand. Actual supply remains subject to product readiness, wafer-start schedules and contractual terms.
Addressing whether customers could transfer their allocated capacity to others with stronger demand, Wei made clear that TSMC does not permit customers to sublease or auction their capacity allocations. In specific circumstances, customers may collaborate through mask licensing agreements, but such arrangements are uncommon and involve only very limited volumes.
TSMC's Terafab talks with Musk hinge on operational control
Elon Musk brings substantial financial resources, internal chip demand across his businesses, and the ability to coordinate assets from Tesla, SpaceX, and other companies. TSMC, meanwhile, possesses advanced semiconductor manufacturing technology, process integration expertise, yield management capabilities, and extensive mass-production experience.
Any collaboration would therefore involve more than financing and site selection. The central question is who would control process management, manufacturing operations, and responsibility for production quality.
Wei confirmed that TSMC has been in contact with representatives of Terafab and dispatched a team to Texas for an assessment. He reiterated that any fab involving TSMC must be operated by TSMC itself. No partnership or investment agreement has been confirmed, and evaluating potential facilities in Texas does not constitute a decision to join Terafab.
Musk intends to combine resources across his businesses through the Terafab semiconductor manufacturing project to meet computing demand associated with AI, robotics, autonomous vehicles, and space applications. Intel joined the initiative in April 2026, with CEO Lip-Bu Tan reaffirming the company's continued participation on October 7.
TSMC's manufacturing expertise, however, cannot be fully replicated in a short period simply through capital investment or equipment purchases. Its insistence on operating any participating fab reflects the company's requirements for process management, production quality, and operational efficiency.
Notably, TSMC has confirmed that it is evaluating options for expanding manufacturing capacity in Texas.
Supply chain sources say the decision will depend not only on customer demand but also on US government policy, water and electricity availability, talent, utilities, government support, and local supply chain conditions.
The assessment could also give TSMC leverage in negotiations with state governments seeking semiconductor investment, allowing it to evaluate independently developed facilities while coordinating suppliers and gradually establishing a regional semiconductor manufacturing cluster.
Terafab could become a potential partner in TSMC's US expansion, but the company does not need to participate in Musk's project to increase its American manufacturing footprint. Existing customer orders, advanced-node capacity expansion, and independently developed fab projects already provide a foundation for further investment.
If TSMC ultimately builds its own Texas fab and develops a supporting local ecosystem, it could establish a semiconductor manufacturing model separate from Terafab.