Citi: TSMC expands multiple process generations at once as demand broadens and overseas costs rise

Why are more fabs needed in this cycle?

TSMC sees demand no longer concentrated mainly on the latest process: N2, N3 and N5 all have demand extending over multiple years. The company expects each major process node to need 8 to 10 fabs, compared with the usual 2 to 4 in the past.
One reason is that chiplet architecture allows customers to combine advanced computing chips with more mature functional chips, benefiting different process generations at the same time. N2 demand is expected to remain strong through the end of the decade; although N3 entered production at the end of 2022, it still needs about 4 additional fabs, while N5 and N7 capacity remains fully utilized.
Management expects AI accelerator revenue to grow at a compound annual rate in the mid-to-high 50% range from 2024 to 2029. Smartphones, non-AI high-performance computing, automotive and the Internet of Things will also contribute to growth as chip content increases and processes are upgraded.

TSMC expects capital spending of 60 billion–64 billion US dollars in 2026, up from about 41 billion US dollars in 2025; cumulative investment over the next three years will also be significantly higher than in the previous three years. Nearly 20 fabs are currently under development.
The company determines capacity using its own demand assessment rather than simply adding up customer forecasts: when customers compete for market share, their combined forecasts may overstate the opportunity. Its assessment covers product road maps, cloud providers’ capital spending, data-center construction, power and financing conditions, with greater weight given to financially sound cloud service providers.
Construction timelines, equipment and engineering talent constrain the pace of expansion. Management considers power the most important potential bottleneck from 2029 to 2031, but recent site visits to US data-center projects indicate no major power constraints from 2026 to 2027.

The start of N3 production in Arizona has been brought forward from the end of 2028 to late summer 2027. The latest process still enters mass production first in Taiwan, while overseas fabs focus more on technologies one or two generations behind to meet customers’ geographic diversification needs.
Overseas fabs are smaller, construction and labor costs are higher, and supply chains are less mature. Management expects dilution of the company’s gross margin from overseas expansion to rise from about 2 percentage points in 2026 to 3–4 percentage points in the following years; overseas costs are still not expected to reach parity with Taiwan within the next five years.
The company hopes to offset cost pressures through pricing, production efficiency, product mix and economies of scale. N2 is making better profitability progress than N3 at a comparable stage of its mass-production ramp, but the initial rise in N2’s share of revenue will still dilute gross margin.

Citi’s model forecasts TSMC revenue growth of 47.6% and 43.2% in 2026 and 2027, respectively, with gross margin rising from 59.9% in 2025 to 66.6% in 2026 and 67.1% in 2027.
Citi has a Buy rating and a target price of 4,000 New Taiwan dollars per share, based on a price-to-earnings multiple of 25 times forecast earnings per share for 2027; the bank believes TSMC’s lead in advanced processes and the AI demand outlook support this valuation.