Taiwan Semiconductor FY26Q2 conference call: Q2 gross margin 67.7% exceeds guidance, full-year capital expenditure raised to $60-64 billion

According to the Zhitong Finance APP, Taiwan Semiconductor (TSM.US) held a FY26Q2 earnings call. The company announced a total cash dividend of TWD 467 billion for 2025, a year-on-year increase of 28.6%, with an annual cash dividend of TWD 18 per share; the cash dividend per share for 2026 will rise to TWD 24, a further increase of 33% year-on-year, with management committing to continue increasing it in 2027.

In terms of performance guidance, Q3 2026 revenue is expected to be between USD 44.6 billion and USD 45.8 billion, with a midpoint corresponding to a quarter-on-quarter growth of 12% and a year-on-year growth of 37%; gross margin is expected to be 65%–67%, and operating margin is expected to be 56%–58%. The company has raised its full-year revenue growth forecast for 2026 in USD terms to "slightly above 40%" (previously over 30%).

Regarding gross margin, Q2 gross margin increased by 150bps quarter-on-quarter to 67.7%, slightly exceeding guidance; the midpoint of Q3 gross margin guidance is expected to decrease by 1.7% quarter-on-quarter to 66%, mainly due to the steep ramp-up of 2nm diluting gross margin by about 3–4%. The ramp-up of 2nm in the second half of the year is expected to dilute gross margin by 3–4%; the ramp-up of overseas fabs is expected to dilute by 2%–3% in the early stages and expand to 3%–4% in later stages.

In terms of capital expenditure, the company has raised its full-year capital budget for 2026 to USD 60 billion–64 billion (previously guided at about USD 52 billion–56 billion in January, and about USD 56 billion in April). About 70%–80% will be used for advanced processes, and about 10%–20% for advanced packaging, testing, etc. Q2 single-quarter capital expenditure was about USD 15.7 billion; cash and cash equivalents at the end of the quarter were TWD 3.5 trillion (about USD 110 billion); the ramp-up of N2 caused inventory days to increase by 7 days quarter-on-quarter to 87 days.

Q&A

Q: Will this upcycle provide capital expenditure guidance for the next three years (26–28) like in 2021?

A: Currently, there are no specific numbers to share. Our capital expenditure this year is aimed at future business opportunities, and we will not hesitate to invest as long as there are business opportunities. As we mentioned in our prepared remarks, we have strong confidence in the long-term AI trend and are increasing capital expenditure, raising this year's capital budget. Last time we mentioned that capital expenditure for the next three years would be significantly higher than in the past three years; the current situation is that capital expenditure for the next three years will grow even more significantly than in the past three years.

Q: After the additional USD 100 billion investment, the cumulative investment in Arizona reaches about USD 265 billion. What are the plans and timelines for capacity deployment in Arizona in the coming years?

A: The timeline depends on market conditions. Given the current situation, the trend is very strong, so we announced additional investment in Arizona. How many factories will be built? Many. Specifically, it is likely that about 4 more fabs will be newly constructed. (Confirmed to include previous funding arrangements.)

Q: How do you respond to competition from wafer foundry rivals like Samsung (which gains high profits from its memory business) and Intel (which receives support from U.S. policies)? Some U.S. customers are engaging with these competitors, and ASML has announced plans to expand EUV capacity by 2028. Are you concerned about competitors seizing more capacity and competing with you? A: Yes, I have a competitor in South Korea who has made a lot of money, and I envy them; another competitor in the United States has received strong support from the U.S. government—by the way, we also received government support, just not publicly announced. But as we say, there are no shortcuts. This means that in the semiconductor industry, we must return to fundamentals. We welcome and appreciate government support, and money is certainly important. But the most important things are always technology, manufacturing, and customer trust; this fundamental has never changed in my 37, 40 years of career, and it has always been the secret to TSMC winning business. From a competitive perspective, choosing a technology and scaling it up for mass production is not like going to 7-Eleven to buy milk—it's not that you see a better bottle of milk today and then go to the next store, and if you don't like it, you switch again. You really have to use it, validate it with test chips, work collaboratively, and prepare capacity to scale up, which takes about 5 years.

Q: Since you see strong signals released by customers and their customers and have raised the full-year guidance, will you revise the 5-year (semiconductor/AI) CAGR? Previously it was around the high 50% range; at the same time, the rising storage costs account for a significant portion of AI capital expenditures. How should we view TSMC's contribution to AI-related growth?

A: If you read our information—we are continuously increasing investments and raising capital expenditures, all for good reasons. If you ask about AI's CAGR, I won't give you a number, but the answer is getting stronger and stronger and stronger. The reason I’m not giving a number today is that it is continuously rising, stronger than what we said before.

Q: In the competition for advanced packaging, especially EMT (embedded/local interconnect packaging), which is gaining market attention, how is TSMC responding?

A: Our packaging capacity is very tight, which has limited customer growth. A bit more flexibility in the market would actually help TSMC's wafer business (which is the main part of TSMC's business). This technology reportedly looks good, and we hope it succeeds to take on some of TSMC's orders. We are currently working hard to narrow the gap between demand and capacity, so we welcome more alternatives to provide flexibility for our customers.

Q: As a new technology, if problems arise after customers adopt it, how will TSMC handle such special needs, and are there alternative solutions?

A: Our top priority is to support our customers' success. We are willing to do anything that helps our customers' business.

Q: When planning capacity expansion, in addition to the needs of customers and their customers, will you also consider the competitive pressure brought by competitors building capacity (as a market leader, long-term supply shortages are not ideal)? How long does it take to meet current demand? Additionally, besides chip shortages, there are issues like data center delays and power supply; how are these incorporated into the planning framework?

A: Competition is the primary consideration every time we think about business; then we assess our position, using a combination of bottom-up and top-down approaches to evaluate demand, which is the most challenging part of our daily work. We make many judgments and communicate more cautiously with customers and their customers (i.e., CSPs) to gather all demand inputs before making judgments. It’s important to note that I believe every customer has told me the truth, but gathering all customers' The truth is, "adding it all up doesn't necessarily lead to the truth—because all CEOs are very aggressive, that's their job. So we need to make very cautious judgments, which may not be entirely correct, but because this involves a large amount of capital (this year's capital expenditure has been raised from about $52 billion–$56 billion to $60 billion–$64 billion, and will continue to increase), we are being very careful. We are also checking the progress of AI data centers—construction, site selection, demand, rent, etc.—to ensure that TSMC's chips will not be piled up as inventory.

Q: Even with such a large-scale capacity expansion plan, do you still believe that supply will remain short until the end of next year?

A: You want me to give a guarantee. I believe that from now until around 2029 or 2030, demand will be very strong. This trend is so robust that I think we are witnessing a brand new industry—I call it the AI industry, which has penetrated daily life and will affect automobiles, humanoid robots, and all industries. In terms of the scale of investment from everyone (including all CSPs), this is a very important new industry for the world, and demand will always exist, fundamentally based on semiconductor chips, most of which are produced by TSMC.

Q: Regarding profitability: In the long term, the profitability of foundries (especially advanced processes) should be higher than that of memory; now that TSMC is no longer the most profitable semiconductor manufacturer, does this mean that the pressure to pass on and capture its own value to customers is smaller than it was a year ago?

A: Your question is actually very simple—what is TSMC's pricing strategy, and what gross margin should it have? Of course, the higher, the better. But we are partners, and I have said many times that our customers must succeed; I don't want to extract too much value from the market. Moreover, we are a company that customers can trust, and we won't suddenly raise prices by 4 or 5 times, as that would make it impossible for customers to survive. We earn the value we deserve and ensure that profits and gross margins are sufficient to support long-term sustainable expansion, which benefits both customers and TSMC—that's our philosophy. So I do envy the 86% gross margin of memory companies—if I had 68%, I would be very happy. In short, we are a very trustworthy company.

Q: As AI demand significantly surpasses other end markets, the top five customers account for a historically high proportion; how do you view the risk of customer concentration?

A: This is not our concern. Customers are getting bigger, and we are happy about that; some customers are growing very quickly. And it's not just that "big customers are getting bigger"; in fact, many new players have emerged in the AI industry.

Q: Your direct customers are providing financing/investment to their end customers to support AI demand; will TSMC consider making such investments or financing arrangements for the customers of its customers (end customers)?

A: To answer you directly: every company has different considerations and strategies. So far, TSMC does not make such financial arrangements because we work very smoothly and successfully with existing customers in the current model.

Q: How will the additional $100 billion investment in the U.S. be distributed over the next 3–5 years, and what will the pace be? A: We do have plans, but the progress and timeline largely depend on market conditions and customer demand. If you ask me for a definite timeline, I don't have one today, but we have plans and will accelerate the process as much as possible. (Supplement: Our new factory and facilities in Taiwan will also be expedited, and the new factory in Japan will be put into production as soon as possible, as there is currently a significant supply-demand gap, and we are working hard to narrow this gap.)

Q: When will the computing part in HPC (CPO/co-packaged optics, etc.) platforms make a substantial contribution to revenue?

A: Production has already started, and it will gradually ramp up over time. I believe that AI data centers need to reduce power consumption and enhance communication channel bandwidth, so the demand for this technology will continue to grow and will become quite important in the coming years.

Q: Under the growth potential of Agentic AI and CPU, how do you see the growth potential and visibility of different AI chips (GPU accelerators vs CPU/XPU)?

A: I can't provide very specific numbers, but I can share that they all use TSMC and the same advanced processes. We are working with customers to allocate wafer supply to balance the ratio of CPU, GPU, and XPU.

Q: Regarding advanced packaging, a roadmap for 14 times the mask size was previously announced to achieve larger AI packaging, and recently in Japan, the development of CoWoS glass substrates was showcased; how is the progress of new technologies such as glass core, glass substrates, and glass carriers?

A: CoWoS is still the mainstream today, and we are developing alternatives to reduce costs, as well as collaborating with substrate suppliers to help customers bring their products to market. The pooling (related production lines) we announced a few quarters ago is under construction and will take about another year to mature, at which point we can enter mass production with customers.

Q: Can you quantify the sales growth outlook for the next few years? Additionally, please break down the key drivers for this round of capital expenditure increase and this year's demand—does it still mainly come from cloud computing, or is it spreading to edge computing? Are there also factors related to equipment supply chain price increases?

A: Because revenue corresponds to investment—we first forecast demand, make assessments, and then decide on capital expenditure. The next few years will be very good business for TSMC, and that's all I can say. (Key drivers: Everything related to AI.)

Q: The competition in backend packaging is intensifying (especially Intel's EMT); are you worried that TSMC's overall foundry value from manufacturing to packaging will be eroded?

A: The frontend wafer business and backend business are two different things; they are not the same. Moreover, our backend capacity is in a state of shortage, with a larger gap, so I welcome competitors to provide some flexibility for my customers to package frontend wafers, which actually helps TSMC's wafer sales. That's our attitude.

Q: You previously mentioned that High-NA equipment is too expensive; how do customers view the stitching challenges brought by smaller exposure fields? Even with technological improvements, will this slow down the adoption of High-NA?

A: You have a deep understanding of High-NA. Currently, its exposure field is only half the size, and we have taken this into account in manufacturing costs and other considerations. Whether we use it or not, High-NA is a very good tool with excellent performance TSMC has clearly stated its cooperation with ASML, striving to make it more suitable for manufacturing in terms of cost and maturity. We will always comprehensively consider technology maturity and cost before deciding whether to adopt it.

Q: The market generally assumes that the unconstrained demand for 3nm and below exceeds your supply capacity by 30%-50%. Is the actual gap even larger?

A: We do not have specific numbers to share. The gap is significant.

Q: It was mentioned at the technology forum that the capacity of the 2nm family will grow at approximately 17% CAGR from '26 to '28, while N3 and N5 will grow at about 25% CAGR. Are these assumptions still valid today? Has there been any change in the past quarter?

A: The figure we presented at the technology forum—now the numbers are larger, and that's all I can say.

Q: Capital expenditures for advanced packaging have always been bundled with testing, photomask manufacturing, etc., accounting for about 10%-20% of total capital expenditures. What proportion is actually directed towards advanced packaging? How do you view the gap between the revenue proportion from packaging and the capital expenditure proportion? Should it be considered as a separate capital expenditure item?

A: We work very hard to ensure the accuracy of capital expenditure numbers, but flexibility needs to be maintained between the front end and back end. Sometimes when bottlenecks occur, we invest more funds to buy bottleneck equipment, sometimes in the front end, sometimes in the back end. In the long term, this ratio is roughly at the level I have always shared, with the back end accounting for about 10%-20%, and this range itself is quite broad. Honestly, over time, as some customers' products require more testing and testing demand increases, we will invest more capital expenditures in testing machines, packaging, or other areas, so we cannot specifically break down how much is invested in each area.

Q: Since the beginning of the year, TSMC has raised its capital expenditure guidance by nearly $10 billion. Where does the upward adjustment mainly come from? Compared to six months ago, is it due to changes in CPU/accelerator, memory manufacturers, or back-end capacity expansion?

A: The most important reason is the continuous growth in demand. We feel the pressure from customers pushing (more accurately, cooperating with TSMC) for capacity expansion, which is one of the main reasons. The second reason is inflation—now we are purchasing equipment at inflation-adjusted prices.

Q: The market is focusing on AI advanced processes, but there seems to be a strong recovery in demand for mature processes and some supply tightness. How do you view the supply-demand dynamics and pricing of mature processes (which have both AI spillover effects and heavily rely on weak consumer demand)?

A: Mature processes cover many different sub-sectors, and only the parts related to AI are in short supply—most importantly, power management ICs (PMICs), because all AI data centers require a large amount of power management, which belongs to mature process technology and is indeed in short supply; additionally, there are sensors, as a large number of sensors are needed to collect environmental information for AI data centers to analyze. Besides that, as you pointed out, the demand for consumer products has begun to weaken, and the demand in other sub-sectors is not strong, so there is no large-scale shortage