Taiwan Semiconductor Q2 2025 conference call: N3 and N5 capacity is very tight, future demand is very high

According to the Zhitong Finance APP, recently, Taiwan Semiconductor (TSM.US) mentioned in its Q2 2025 earnings call that N3 capacity is very tight and will remain tight for the next few years, while N5 is also tight. Demand is high because many AI applications are still using N4, and will switch to N3 in the next two years. N5 is still capacity-constrained, and N3 capacity is even tighter. Taiwan Semiconductor's advantage lies in having a giga fab cluster, with 85-90% of the same equipment for each process, including N7, N5, N3, and the future N2.

Regarding the semiconductor opportunities for humanoid robots, it is still too early to say that humanoid robots will contribute this year, and it may be too early next year as well. It is expected that robots will be used first in the medical industry, and robots require many sensors. Some customers say the opportunities for robots are ten times that of EVs.

Q&A

Q: Regarding demand, you mentioned that AI demand looks better than before. You previously said COWOS would reach supply-demand balance in 2026. How do you view this now?

A: AI demand is getting stronger, as you can see from what the CEOs of these CSPs are saying. COWOS demand is also strong, and we are in a process of trying to reduce the gap between supply and demand. It’s not about balance as you mentioned, but about narrowing the gap.

Q: What changes have you seen regarding edge AI?

A: We believe it may take one or two years for our customers to complete new designs and new products, and the momentum is still continuing. The growth at the edge and the increase in shipments are actually moderate, but die size will grow by 5-10%. It may take 6-12 months to see an explosion.

Q: Considering the full-year guidance, Q4 revenue is expected to decline. What is the reason?

A: We will strive to achieve high targets. Your calculations are based solely on the numbers we reported, which can lead to a reduction. We have considered potential tariff impacts and other uncertainties, which has made us more conservative; this is our current attitude. Our leadership in advanced processes and manufacturing excellence gives us hope to achieve higher targets whenever there is an opportunity.

Q: Regarding gross margin, when considering pricing for 2026, will the company take exchange rate impacts into account? In the long term, will the company’s future gross margin reach a target better than 53%? Will next year’s gross margin be similar to this year’s?

A: The impact of exchange rates is significant. As for whether we will consider selling value during the pricing period, we are indeed working on it, and we expect a gross margin higher than 53%. We hope you will pay attention to this higher figure.

Q: Regarding the restoration of H20 sales to China, after the last H20 ban, you were still confident in achieving a 45% CAGR over five years. Do you now believe the CAGR will be higher?

A: We have not seen more signals regarding H20, but it is indeed a good thing. China is a large market, and this is very positive news for us. We do not think it is the right time to raise guidance yet; perhaps next quarter Q: How much will N2 contribute to revenue? Will it reach a 15% share next year?

A: Generally, when a new process is ramping up, it is primarily used in smartphones. But now it's not just smartphones; there are also HPC products. However, the ramping is similar to N3, mainly because our new fab construction is somewhat limited, and capacity is constrained. We believe the ramping is similar, but the revenue contribution is better because the pricing is different.

Q: Will the growth rate of N2 be very fast in 2027?

A: We will answer this question in 2026.

Q: What is the supply and demand situation for N3 and N5 in the next two years? Many products are moving to N3 next year, and the transition from N5 to N3 is basically over, but we haven't seen new N3 factories being built. Will supply and demand be very tight? Will there be more N3 expansions in the future? Will N3 and N5 sell out?

A: Indeed, the capacity for N3 is very tight. I really like the idea of selling our value; it will be tight for the next few years. N5 is also tight. Demand is high because many AI applications are still using N4, and they will switch to N3 in the next two years. N5 is still capacity-constrained, and N3 capacity is even tighter. Our advantage is that we have a giga fab cluster, including N7, N5, N3, and the future N2. We have 85-90% of the same equipment for each process, so it's relatively simple for us to switch capacity. We are using N7 capacity to support N5 because N5 is too tight. As N5 transitions to N3, we will continue to do so. We define that N7 and more advanced processes are all very tight. We are working hard to reduce the supply-demand gap.

Q: Regarding gross margin, will the company offset the dilution of gross margin for overseas factories through pricing? NV mentioned that they are collaborating with the company to calculate lithography efficiency improvements. Can you share some ways to improve costs?

A: Regarding gross margin, we say there are six factors that affect our gross margin. When exchange rates fluctuate, for example, we encountered similar situations a few years ago, and we relied on other factors to help us mitigate some of the negative impacts, allowing us to still meet our gross margin targets. Pricing is just one of the aspects; there are other factors we can improve as well. We say that 53% AND HIGHER is achievable. Regarding the benefits of AI, we have mentioned before that we use it in manufacturing, operations, and R&D. If we can achieve a 1% improvement in production efficiency, it means $1 billion for us.

Q: You mentioned plans to build 11 factories in Taiwan and 8 factories elsewhere. This is a significant project. Will there be plans for more capacity expansion next year? Recently, many CSPs have announced new data centers, and demand is strong. Will the company have enough capacity to meet next year's demand? Will more N5 migrate to N3?

A: Indeed, we have recently seen many announcements regarding AI data centers. The demand for N3, N5, and future N2 is very strong; we haven't seen such good demand in a long time, and we are still working to reduce the supply-demand gap Q: What is the ROI comparison between N2 and N3? What are the CAPEX and yield for N2?

A: In terms of N2's return rate, compared to N3, it is like the old days achieving the company's gross margin, and back then the gross margin was still 50%, while now we are talking about a gross margin of 53% or higher. N2 indeed has higher profitability compared to N3. We expect mass production in H2 this year, and revenue will start to pick up in H1 next year.

Q: The company's CAPEX guidance has not changed; is this a conservative stance on CAPEX? Is it due to uncertainty? Will CAPEX accelerate in 2026 and 2027?

A: The CAPEX this year reflects the opportunities we see for the second year. If we see opportunities, we will not hesitate to invest. We are aware of the macro uncertainties, and we will consider these. Looking ahead, it is too early to talk about investments for the next few years. However, for a company of our size, we should not see a sudden significant drop in CAPEX.

Q: Is there any adjustment to the growth rate of AI revenue for 2025? Will COWOS capacity increase?

A: We are working to narrow the gap, and it is similar now and in 2026. Demand and momentum are both very healthy and strong. We are building a lot of capacity in backend to support our customers. The demand for AI is very strong. The demand for COWOS is also very strong.

Q: AI chips are getting larger and consuming more power. Besides advanced processes, the company has also announced some advanced packaging technologies. What is the timeline for these new technologies?

A: TSMC's technology development is based on customer demand; we will only expand production if there is customer demand. Every customer is very important to us. In the advanced packaging field, many customers adopt different approaches, and we have developed many different technologies. We have a systematic integration solution, such as COWOS and other technologies, to support our customers.

Q: Can these technologies be easily converted?

A: Indeed, there are many similarities between the technologies; otherwise, we would have put in too much effort.

Q: We see strong demand for advanced support and advanced packaging, but there is an oversupply in mature processes. What is the company's strategy for mature processes?

A: There are many mature processes now. Our strategy for mature processes is to develop specialized processes. For example, RF, high voltage, sensors, etc., we base our development on customer demand. We are not overly concerned about oversupply; if there is oversupply, we will not build factories in Japan and Germany. It all depends on customer demand, and these are all specialized processes.

Q: Humanoid robots are already in use at the company. How do you view the semiconductor opportunities for robots?

A: It is still too early to say that humanoid robots will contribute this year, and it may be too early next year as well. I expect robots to be used first in the medical industry; robots require many sensors. Some of our customers say that EVs are nothing compared to the opportunities presented by robots, which are ten times that of EVs.

Q: Due to pricing reasons, will customers pull goods in advance before 2026? A: No, we haven't seen any abnormal behavior from customers. Although the demand for N3 is strong, it requires manufacturing time, and customers cannot pull goods in advance; the production capacity is very tight. We only have a small space for customers to pull goods in advance.

Q: In the past, when investing in N3 and N5, the company's capital expenditure intensity was above 40%. Now that the demand for N2 is strong, will the capital expenditure intensity return to the previous level of 40% or even higher?

A: The capital expenditure intensity is related to future growth opportunities. If we make investments and see rapid revenue growth, CAPEX will also increase, without needing a high capital expenditure intensity. We do not set a target for capital expenditure intensity; it is related to future opportunities, and this is not that meaningful.

Q: The A16 is mainly used for HPC; will AI be the first to use it?

A: Indeed, HPC customers generally use the previous generation of processes. Now, due to strong AI demand, energy consumption is critical. The A16 has significant energy efficiency improvements, which are very important for AI customers. Our customers are migrating faster. Due to energy consumption, end customers need a large power plant; the A16 represents further improvements compared to N2, so we are not surprised that we expect AI industry customers to use the A16.

Q: The strong demand for the second factory in the U.S. has accelerated expansion. Now that the new U.S. legislation provides tax reductions for CAPEX, what impact does this have on the company?

A: Expansion is driven by customer demand. We appreciate the U.S. government's legislation, but the actual acceleration of expansion is caused by customer demand. The gross margin aspect is positive, but it won't be significant within five years.

Q: The expansion in the U.S. has accelerated; what about expansion in other regions? Will future CAPEX differentiate overseas capacity?

A: The U.S. focuses on advanced processes, Japan on specialty processes, mainly for CMOS, and Germany primarily for automotive use. These are not in the same field; investments in the U.S. or advanced process investments will not affect investments in Japan and Germany.

This article is sourced from the "Guojin Electronics Research" WeChat public account, edited by Chen Xiaoyi of Zhitong Finance