TSMC Confident in 56%+ Long-Term Gross Margin

According to TSMC Chief Financial Officer Wendell Huang, the company remains confident that its long-term gross margin can stay above 56%, despite near-term pressure from the ramp-up of 2nm mass production and overseas capacity expansion. He also said AI is still in the early stages of development, with AI agents expected to broaden computing demand from accelerators to processors. Meanwhile, TSMC’s CoWoS advanced packaging capacity remains tight, prompting the company to outsource part of its packaging processes to ease capacity constraints.

TSMC recently participated in Goldman Sachs’ Communacopia + Technology Conference 2026 in San Francisco, where Huang discussed AI demand, advanced process technologies, advanced packaging, overseas expansion, and profitability with investors.

Huang noted that demand for TSMC’s 3nm (N3) process is following a different pattern from previous advanced-node cycles. In the past, demand for a new process node typically weakened two or three years after its introduction. However, demand for N3 remains strong, and TSMC continues to expand capacity as supply has yet to fully meet customer requirements.

The situation is similarly tight for CoWoS advanced packaging. In addition to expanding its own capacity, TSMC is working with outsourced semiconductor assembly and test (OSAT) partners to handle part of the packaging process. This approach is intended to prevent backend packaging constraints from limiting shipments of advanced-process wafers.

Huang remains optimistic about the medium- and long-term outlook for AI, saying the industry is still in the early stages of a multi-year growth cycle. As AI evolves from generative AI toward AI agents, computing demand is expected to expand beyond GPUs and customized AI accelerators to CPUs and networking chips. This trend could broaden the range of AI-related applications benefiting TSMC across x86, Arm, and RISC-V architectures.

TSMC also believes that the growth of open-source AI models and improvements in computing efficiency will continue to reduce the cost per token. Lower AI usage costs could accelerate adoption, increasing overall usage and driving additional demand for computing capacity. TSMC has incorporated this trend into its capacity planning. Recent discussions with U.S. cloud service providers (CSPs) have also strengthened the company’s confidence in long-term AI demand, as customers continue preparing power infrastructure and data center capacity for future AI deployments.

Regarding its overseas expansion strategy, TSMC said that under its current plans, approximately 30% of its 2nm and more advanced process capacity is expected to be located outside Taiwan over the next five years, with a significant portion centered on its facilities in Arizona.