TSMC 3nm Surge: Q3 Revenue to Top NT$400B

According to reports, surging demand for AI and high-performance computing (HPC) applications is prompting TSMC to expand capacity for advanced process nodes below 3nm. Industry analysts expect TSMC’s 3nm process revenue to surpass 5nm revenue for the first time in the second half of this year, with third-quarter 3nm revenue estimated to exceed NT$400 billion, setting a new record and accounting for more than 30% of total revenue. The stronger contribution from 3nm is also expected to support TSMC’s gross margin performance.

TSMC’s 3nm process family continues to face strong demand. TSMC Chairman and CEO C.C. Wei previously said during an earnings call that the company had traditionally avoided adding capacity after a process technology reached its target production level. However, the rapid growth of AI applications has changed the situation, leading TSMC to further expand 3nm capacity. Since building a new fab typically takes two to three years, capacity constraints are expected to persist.

In addition to constructing new fabs, TSMC is accelerating the expansion of 3nm production capacity at existing facilities. The company is reportedly converting some 5nm equipment to support 3nm production, allowing greater flexibility across the 7nm, 5nm, and 3nm process nodes. This cross-node capacity optimization is expected to improve production efficiency, while the gross margin of the 3nm process could exceed TSMC’s company-wide average in the second half of the year.

As TSMC continues to expand 3nm production, 3nm-related revenue is expected to exceed 5nm revenue for the first time in the third quarter. At the same time, the company’s 2nm process is reportedly ramping faster than initially expected in its first year of production, providing an additional boost to overall revenue growth.

Beyond 3nm, TSMC is also accelerating the development of next-generation angstrom-class process technologies. Multiple advanced nodes are moving toward mass production or production readiness. The first angstrom-generation process, A16, is expected to become production-ready in the second half of this year, with industry sources suggesting that Fab 20 could be among the first facilities to manufacture chips using the technology.

Industry reports indicate that early A16 customers could include Apple, NVIDIA, and other major AI companies. OpenAI is also reportedly expected to place orders with TSMC for an ASIC developed by Broadcom, potentially making it another important A16 customer.

According to previously released TSMC information, the company plans to begin mass production of its A14 process in 2028. An extended A13 process is expected to enter mass production in 2029, while the second-generation backside power delivery solution, A12, is also scheduled for mass production in the same year.