Taiwan Semiconductor Manufacturing Co Ltd ($TSM) Q2 2025 Earnings Call

Thank you, Wendell. Good afternoon, everyone. First, let me start with our near-term demand outlook. We concluded our second quarter with revenue of $30.1 billion in US dollar, above our guidance in US dollar term, mainly due to continued robust AI and HPC related demand. Moving into the third quarter 2025, we expect our business to be driven by strong demand for our leading edge process technologies. Looking into the second half of 2025, we have not seen any change in our customers' behavior so far. However, we understand there are uncertainties and risk from the potential impact of tariff policies, especially on consumer related and price sensitive end market segments. While we observe rebate programs in China are stimulating some near-term demand upside, we believe this is short-term in nature and continue to expect a mild recovery in overall non-AI end market segments in 2025. Having said that, we believe the demand for semiconductor is very fundamental and will continue to be robust. Recent developments are also positive to AI's long-term demand outlook. The explosive growth in token volume demonstrates increasing AI model usage and adoption, which means more and more computation is needed, leading to more leading AI silicon demand. We also see AI demand continuing to be strong, including rising demand from sovereign AI. Therefore, we now expect our full-year 2025 revenue to increase by around 30% in US dollar term, supported by strong demand for our industry-leading 3-nanometer and 5-nanometer technologies, underpinned by growth in our HPC platform. Amidst the uncertainties, we will remain mindful of potential tariff-related impact and be prudent in our business planning going into H2 2025 and 2026, while continuing to invest for the future mega trend. We are also focused on the fundamentals of our business: technology leadership, manufacturing excellence, and customer trust to further strengthen our competitive position. Next, let me talk about TSMC's global manufacturing footprint update. All our overseas decisions are based on customers' need and the value, some job, graphic flexibility, and necessary level of government support. This is also to maximize the value of our shareholders. With strong collaboration and support from our leading US customers and US federal, state, and city government, we announce our intention to invest a total of $165 billion in advanced semiconductor manufacturing in the United States. This expansion includes plans for six advanced wafer manufacturing fabs in Arizona, two advanced packaging fabs, and a major R&D center to support strong multi-year demand from our customers. Our first fab in Arizona has successfully entered high-volume production in Q4 2024 utilizing N4 process technology with yield comparable to our fab in Taiwan. Construction of our second fab, which will utilize 3-nanometer process technology, is already complete. We are seeing strong interest from our leading US customers and are working on speeding up volume production by several quarters to support their needs. Construction of our third fab, which will utilize 2-nanometer and A16 process technologies, has already begun, and we are looking to speed up production as well based on strong AI-related demand from customers. Our fourth fab will utilize N2 and A16, and our fifth and sixth fabs will use even more advanced technologies. Construction and ramp schedule will be based on customers' need. Our expansion plan will enable TSMC to scale up to a gigafab cluster in Arizona to support leading-edge customers in smartphone, AI, and HPC applications. We also plan to build two new advanced packaging facilities and an R&D center to complete the AI supply chain. After completion, around 30% of our 2-nanometer and more advanced capacity will be located in Arizona, creating an independent leading-edge semiconductor manufacturing cluster in the US. Thus, TSMC will continue to play a critical role in enabling customers' success and maintain a key partner and enabler of the US semiconductor industry. Next, in Japan, thanks to strong support from central, prefecture, and local government, our first specialty fab in Kumamoto has already started volume production in late 2024 with very good yield. Construction of our second specialty fab is scheduled to start later this year subject to local infrastructure readiness. Ramp schedule will be based on customer need and market conditions. In Europe, we have received strong commitment from the European Commission and German federal and city governments and are progressing smoothly with plans to build a specialty fab in Dresden, Germany. The ramp schedule will be based on customer need and market conditions. In Taiwan, with support from the Taiwan government, we plan to build 11 wafer manufacturing fabs and four advanced packaging facilities over the next several years. We are preparing for multiple phases of 2-nanometer fabs in Hsinchu and Kaohsiung Science Park to support strong structural demand from customers. By expanding our global footprint while continuing to invest in Taiwan, TSMC can continue to be the trusted technology and capacity provider of the global IC industry for years to come while delivering profitable growth for shareholders. Now, let me talk about our N2 and A16 status. Our N2 and A16 technologies lead the industry in addressing insatiable demand for energy-efficient computing, and almost all innovators are working with TSMC. We expect the number of new tape-outs for 2-nanometer in the first two years to be higher than both 3-nanometer and 5-nanometer in their first two years, fueled by both smartphone and HPC applications, and deliver full node performance and power benefit with 10-15% speed improvement at same power or 20-30% power improvement at same speed, and more than 15% chip density increase compared with N3E. N2 is well on track for volume production in H2 2025 as scheduled, with a ramp profile similar to N3. We also introduce N2P as an extension of our N2 family, with further performance and power benefits, volume production scheduled for H2 2026. We also introduce A16 featuring our best-in-class super power rail (SPR). Compared with N2P, A16 provides 8-10% speed improvement at same power or 15-20% power improvement at same speed, and additional 7-10% chip density gain. A16 is best suited for specific HPC products. Volume production is on track for H2 2026. We believe N2, N2P, A16, and derivatives will make the N2 family another large and long-lasting node for TSMC. Finally, let me talk about A14 status. Featuring second-generation nanosheet transistor structure, A14 will deliver another full node stride from N2 with performance and power benefits. Compared with N2, A14 will provide 10-15% speed improvement at same power or 20-30% power improvement at same speed, and about 20% chip density gain. Our A14 technology development is on track and progressing well, with device performance and yield on or ahead of schedule. Volume production is scheduled for 2028. We will continue our strategy of continuous enhancement with A14, including a super power rail offering planned for 2029. We believe A14 and its derivatives will further extend our technology leadership position and enable TSMC to capture growth opportunities well into the future. This concludes our key message and thank you for your attention.

Goku, the demand for the AI getting stronger and stronger. If you pay attention to what the four leading companies' CEOs said. And so the macro trend for AI continues to be strong. And so is CoWoS. We are again in a mode to try to narrow the gap. I don't want to use 'balance' because last time you misunderstood what I said. That was bad wording. So I will say we try to narrow the gap. All right. So momentum is still there and very healthy.

As I said last time, it takes one to two years for my customers to complete new designs on the product. The momentum is still going. They continue to increase, but the number of units on edge devices is actually mild. However, the die size increase continues. We see the die size increase by about 5 to 10%. That trend continues. So you have to wait around probably 6 months to 1 year to see an explosion.

I think you misinterpreted Gokul's comment. You are saying we are becoming conservative. Our company is more realistic. We are a company that says what we will achieve and then achieves the high target. So your calculation, I think Charlie also calculated, shows that our fourth quarter is decreasing. We take into consideration the possible impact of tariffs and many other uncertainties. So we become more conservative. That's our current attitude. But I guarantee you with our technology leadership position and excellent manufacturing, if there are any opportunities, we will catch them and expect to achieve our high-end target. Okay.

Well, let me assure you that yes, the impact of the exchange rate is huge. But you try to imply whether we are still earning our value. Let me answer that. We are working on it. And we have confidence that the 53% gross margin and higher — I still want you guys to pay more attention to 'and higher.' Okay. Thank you.

Charlie, regarding the H20, according to the leading companies' CEOs, we have not received the signal yet. So it's too early to give an estimate. But certainly it is good news, right? China is a big market, and our customer can continue to supply chips to that big market. That is very positive news for them, and in return very positive for TSMC. Whether we are ready to increase our forecast? Not yet. Another quarter will be more appropriate to answer your question.

Bruce, you have a good argument. Usually we ramp a new node using smartphones. You knew that. Now it's not only smartphone but also HPC products. However, the ramp profile I just reported says similar to 3-nanometer because it is limited by our capability to build new fabs. Also, it is straightforwardly constrained by capacity. So the ramp profile is similar to N3. But the revenue contribution will certainly be bigger. Because you don't expect N2 to be the same price as N3, right?

Good. Thank you.

We will answer that question in 2026.

I like your comment on that we have to share our value because of very tight in N3 capacity. It will be continued for a couple of years. Very tight. And in fact, N5 also very tight. The demand is high because of a lot of AI product still in the 4 nanometer technology node and they will transition to 3 nanometer probably in next 2 years. So, in meanwhile, N5 are still very tight in capacity. N3 even tighter. And so, we are working hard. One of TSMC's advantages is we have gigabyte cluster. And so, we have between N7, N5, N3, even the future N2, we have almost for each node, we have about 85 to 90% common tools. So, it's not free, but it's much easier for TSMC to adjust or convert the capacity between those nodes. And today, let me share with you, we are using the N7 capacity to support N5 because N5 is too tight. And then we are converting N5 to N3 as you just pointed out. We will continue to do that. And so, today our leading edge technology capacity... We define N7 and below are all very tight. And seeing that, we are working very hard to, again, using my sentence, narrow the gap between the demand and the capacity.

Your observation is right. Recently we saw a lot of announcements of AI data centers all over the world and the demand on 3 nanometer, actually on 5 nanometer, 3 nanometer, and the future 2 nanometer are very high. We did not see this kind of strong demand for a long time. But, will we have enough to support them? I still want to use my word say that we try very hard to narrow the gap between the demand and the supply. We're working very hard.

Well, my answer stays the same. We are trying very hard to narrow the gap. For now, you know, for 2026, the demand, the momentum are very healthy and very strong. And so, we are building many new facilities in the back end to increase the AI, increase the core capacity to support our customer. AI demand is very strong. And so, the core capacity, the demand is very strong.

I think TSMC's philosophy to develop technology is working with customer. The customer has such demand, we develop the technology, we increase the capacity for them. So, priorities, every customer is important to TSMC. And in the advanced packaging side, a lot of customers are using different approaches. So, we are developing a variety of different back end packaging advanced technology for all the customers. Whether it is related to the advanced leading edge technology, the answer is yes. Okay. So, we have system integration. We have CoWoS arrow. That's a terminology. We have a lot of different names that I cannot even remember, but there are a lot of varieties and we work with our customer to meet their demand. That I can answer you.

Of course, there are some similarities in between. Otherwise, we are going to take too much effort and then not get the return. Yes, there are a lot of similarities, but there are a lot of varieties also.

Good question. If you read the newspapers, there is so much mature node capacity. But TSMC's strategy actually is, on the mature node technologies, we develop kind of specialties. For example, that's RF technology or CMOS image sensor or the high voltage. So, we develop the technology at the request of our customer. So, we don't worry too much about what you say, the overcapacity. If it is really overcapacity, we will not build a fab in Japan. We will not build a fab in Germany. So, it's not overcapacity. It's all related to customers' need, customers' demand, and those are all specialty technologies. Did I answer your question?

Brad, it's too early. Actually, it's too early to say the humanoid robot will play a role this year. Next year, probably still too early because it's so complicated. You know that the humanoid robot most of the time will be used... I think the first one will be used in the medical industry to take care of people getting old like me. And you know, I probably someday later I needed some humanoid robot to help me. But, you know, it's very complicated because we are talking about the brain only. Actually, you are talking about a lot of sensor technology. That's the image sensor, the pressure sensor, the temperature sensor, and all the feedback to the CPU. And so, it's very complicated. And since it's dealing with human being directly, it has to be very, very careful. But then, once it starts to fly, it will be a big, big plus. I talked to one of my customers, and he said, 'Sir, the EV car is nothing. The robot will be 10 times of that. I'm waiting for that.' Okay. Did I answer your question?

Well, the answer is no. We did not see any different customer behaviors so far. But let me share with you, I add more color. If you are talking about the 3 nanometer demand for example, the cycle time itself takes about four months. So there's no way you can pull in anything. I mean, that's a, as I said, our capacity is very, very tight. So we already have all the schedules and so pretty little room for pulling, let me say that. Even if they wanted, but no. Answer is no. So 2026 is 2026, we will share with you. Yeah.

I don't know how to answer that because so far AI has been N plus one, N plus two. Is that node A16 and the third node where AI would move to the leading edge?

Well, Madi, you are right. Usually the HPC customer is always one step behind. They're using N plus one or N plus two technologies. Now, because of AI demand is so strong, that's one thing, but the most important thing is they need some kind of performance, but the power consumption is very, very important. And when we talk about A16, we have another power efficiency improvement of close to 20%. That's a big value for all the AI data center applications. So, that helps my customer move faster because every time when we talk about the AI data center, if you notice, the first thing they talk about is power supply, electricity, right? So, they don't tell you that the power efficiency is very important, but they tell you that we have to build a very big electricity power plant to support the AI data center. So, that tells you how important it is. And TSMC's technology, by the way, on A16 is a further improvement of the N2 node. So, it's not a surprise for TSMC to expect for those people in AI data center industry, they want to use A16. Okay.

Okay, let me share with you about our ramp up schedule. It's totally because of our customers' demand. And we appreciate the US government increase the ITC from 25% to 35%. We appreciate that. It helps. But the real schedule is because of our customers' demand. So, we have to prepare the capacity to meet the demand. That's the number one consideration.

Well, you think about TSMC expansion overseas. In the US, it's leading edge. In Japan, it's on specialty technology. To be specific, most of the time is for the CMOS image sensor. For Germany, it's automotive industry. So, they are all not in the same field. So, actually, the investment in the US or investment on the leading edge does not affect the investment in Japan or in Germany.