Source: Wall Street See, Author: Li Xiaoyin

Wei Zhejia, director and CEO of Taiwan Semiconductor, said that the main reason for raising the annual capital expenditure guidance is to consider higher growth opportunities. The part of the revenue that exceeds the guidance in this quarter mainly benefits from the demand for N3 and N5 processes.

Benefiting from the surge in demand for AI, Taiwan Semiconductor's revenue and profit doubled in the second quarter and its advanced process performance was strong. In terms of profit guidance that the market is concerned about, Taiwan Semiconductor raised its sales expectations for the third quarter and the full year, with expected sales of USD22.4 billion to USD23.2 billion in the third quarter and a sales growth rate of 24%-26% for the full year in USD terms. At the same time, Taiwan Semiconductor also raised the lower limit of its capital expenditure for the year from USD28 billion-USD32 billion to USD30 billion to USD32 billion. During the earnings call, Taiwan Semiconductor's Chairman and President Wei Zhejia gave a response to questions on each business performance, CoWoS and next-generation process capacity, capital expenditure expectations, and pricing strategies, among other issues.Wei Zhejia said that the reason why the capital expenditure for the entire year in 2024 was raised to USD30 billion to USD32 billion was mainly due to higher growth opportunities, of which 70%-80% will be used for advanced processes, 10%-20% for specialty processes, and 10% for advanced packaging, masks, and the like.Wei Zhejia also pointed out that over the past three months, strong AI and smart phone support have driven demand, and it is expected that the 5nm process will remain at a high level in the second half of the year. The entire year is expected to be "our strong year of growth, with revenue growth expected to slightly exceed the middle range of 20% in USD terms."

In terms of the industry as a whole, Taiwan Semiconductor expects the annual growth rate of the foundry industry to be close to 10% this year, and last year Taiwan Semiconductor's market share was about 28%.

Wei Zhejia said that with the continuous strong demand for AI-related businesses, we need more advanced processes and advanced packaging. Our technological position will increase accordingly, and we are very optimistic about our long-term development prospects.

It will take at least three years for N3 to reach the company's gross margin.

Taiwan Semiconductor claims that N3, as the company's most advanced chip-making process technology currently, will take longer to achieve the company's gross margin, possibly taking 10-12 quarters. As for the more advanced 2nm process technology, Wei Zhejia said that the early production of 2nm will be higher than that of 3nm and 5nm, with device performance improving to 25%-30% and chip density increasing by more than 15%. It is expected that by 2026, the revenue from the 2nm process will exceed that of the 3nm. It is expected that N2P (performance-enhanced 2nm) will be mass-produced in the second half of 2026.

Regarding CoWoS production capacity, Wei Zhejia responded that the current demand for CoWoS production capacity is still strong and it is difficult to meet customer demand. The tight supply situation is expected to continue until 2025 and may ease in 2026. It is expected that the company's CoWoS production capacity for this year and next year will double at least.

In terms of pricing strategy, Taiwan Semiconductor has always emphasized the "sell the value" strategy, and stated that in order to achieve the long-term gross margin target of reaching 53% or higher, it may attempt to shift costs to customers.

Wei Zhejia reiterated the "sell the value" strategy during the call and did not currently raise the gross margin target of 53%. He said that "after we communicate more with customers, we may inform them of specific higher rates."

The following is a summary of the Taiwan Semiconductor Q2 earnings call by the SinoLink Electronic Research Team:

Due to the surge in demand for AI, gross margin for the N3 process may take a minimum of three years to achieve.

Taiwan Semiconductor claims that the N3 process, as the company's most advanced chip-making process technology currently, will take 10-12 quarters to achieve the company's gross margin.

Wei Zhejia said that the early production of the 2nm process technology will be higher than that of the 3nm and 5nm, with device performance improving by 25%-30% and the chip density increasing by over 15%. It is expected that by 2026, the 2nm process technology's revenue will exceed that of the 3nm process technology. It is expected that N2P (performance-enhanced 2nm) will be mass-produced in the second half of 2026.

Wei Zhejia responded that current CoWoS production capacity demand is still strong and it is difficult to meet customer demand. The tight supply situation is expected to continue until 2025, and may ease in 2026. The company's CoWoS production capacity is expected to double at least for this year and next year.

Taiwan Semiconductor has always emphasized the "sell the value" strategy and stated that in order to achieve the long-term gross margin target of reaching 53% or higher, it may attempt to shift costs to customers.

Taiwan Semiconductor reiterated the "sell the value" strategy during the call and will not raise the gross margin target of 53% for the time being. It stated that "after we communicate more with customers, we may inform them of specific higher rates."

The following is a summary of the Q&A session of the Taiwan Semiconductor Q2 earnings call by the SinoLink Electronic Research Team:

Q: Looking at future production capacity planning, there is a lot of demand for advanced packaging for AI-related businesses. How will this production capacity be planned in the future? What about CoWoS production capacity in the next year?

Demand is strong, and it is difficult to meet customer needs. It is not easy to achieve balance until 2025 or 2026, and the CAGR of production capacity cannot be predicted. Supply will remain tight until 2025, and it is hoped that it can be eased in 2026.

Q: Will the production capacity double next year?

We said it would double this year, now it is more than double. If we say it will double next year, it is estimated that it will more than double again. Anyway, we are working very hard.

Regarding gross margin, H2's gross margin is better than expected. Looking at the rising gross margin, how will the company's gross margin look in the next few years under the sell value situation? Will it reach high 50s?

Gross margin has positive and negative effects. The positive aspects are N3 climbing dilution, sell value, and cost reduction, which we are good at. In addition, for example, the conversion from N5 to N3, we will not rule out the possibility of more conversion in the future. We see strong demand for N3 and this will have a negative impact in the future. However, this is good for the next few years. There are also effects such as electricity prices. We are also beginning to invest in overseas factories. There will be two next year, phase 1 in Arizona and a factory in Japan, which will dilute our gross margin by 2-3% in the next few years. However, overall, we have confidence in achieving a gross margin of 53% or even higher with our ability to manage costs and factories. It is possible to achieve the gross margin in 2022 if we achieve high utilization rates.

Will different government incentives and subsidies affect the company's CAPEX?

When we receive subsidies, we see them on the cash flow statement and balance sheet. Different governments have different methods. In our financial reports, we have received subsidies in the past few quarters. For example, in 2023, we received subsidies slightly higher than $1.5 billion, mainly from Japan.

Regarding sell the value, what is the current situation? Will advanced processes become a bottleneck in the future and allow the company to sell value more effectively?

Pricing strategy is very strategic and the current situation is still very good. This is a continuous process, as we continuously sell value, and our customers are doing well, so we should be able to do well too.

HPC customers are doing well, but will cost-sensitive smart phone customers react differently to price increases?

Pricing is strategic, and not the same for every type of product. Our customers are looking for advanced process capacity, and we work with them to support them in both price and capacity.

Regarding Trump's comments, how will the company respond? Will it expand overseas production capacity?

So far, we have not made any changes to our overseas expansion plans. We will continue to expand in Arizona and Japan, and may also expand in Europe in the future. If tariffs increase, the customer will be responsible.

The company has been emphasizing 'sell the value' recently. Why has the long-term gross margin not been raised as a target?

We work with our customers and pricing is a strategic issue. Of course, we want to sell value. If we change our gross margin target now, we will emphasize 53% or even higher, and we will not change the specific number now. When we have more communication with our customers, we may inform them of the specific higher magnitude.

Regarding advanced packaging, the margin of advanced packaging is lower than the company's average. Will Cowos' gross margin be higher? Will the company cooperate with more partners to provide Cowos supply?

All customers do want to migrate to better efficiency processes and reduce power consumption. Many customers want to move quickly. We are building capacity and it is tight right now. We hope that in the next few years we will be able to support the demand. At present, we are indeed making efforts to support them.

Regarding the company's next generation process capacity, AI customers are eager to move to the most advanced support, mainly due to energy consumption. Will N2 and A16 be larger processes than previous processes?

Indeed, all customers hope to migrate to better energy-efficient processes and reduce power consumption. Many customers want to move quickly. We are working hard to build capacity, and it is still very scarce at present. We hope to be able to support the demand in the next few years.

Regarding gross margin, what is the current situation for N3? Has N3E raised the gross margin of N3?

We do not differentiate between gross margins of the same process. Generally speaking, N3 needs a longer time to achieve the gross margin of the company, it used to be 8-10 quarters, and N3 may take 10-12 quarters, but it is now improving and we expect it to continue to improve.

From the perspective of the company's process conversion strategy, the company has always emphasized process flexibility and has done so in the past. Are N5 and N3 a big process?

N12 and N16 are indeed a big family, but N5 and N3 are not. At the same time, the similarity to the process is relatively high, with a 90% similarity between N5 and N3, and both are in Tainan, so it is easy to migrate production capacity.

Regarding Cowos, will the company's Cowos capacity double in 25 years or more? Will customer migration from Cowos-S to Cowos-L, where L/R does not require TSV and large Interposer, help ease the shortage of capacity? When will it reach balance in 25, 26?

Cowos-L/R/S are based on customer demand, and different versions of cowos are added together. It's not about sharing which version is doubling or has more points. We don't just double, we hope to double this year and next year, and we need to work with all our partners to support our customers. Different versions of cowos require different tool sets, and some tools can be used by all versions, but different versions still have different requirements.

Regarding advanced processes, looking at the migration of the 2nm process, will the revenue share of 2nm in 26 be larger than N3? Is the dilution of N2's gross margin also shorter than that of N3?

The revenue will be larger. The dilution of the gross profit margin will reach the company's average gross profit margin faster.

Regarding packaging, have clients already begun to use 3D SOIC and similar packages on the terminal side of AI? Will more smart phone customers start using info first packaging?

Our customers enter N2 and A16, they need to use a chiplet solution, which requires advanced packaging. HPC customers migrate faster and have higher requirements for delay and other issues, while mobile phone customers have demand for footprint and function upgrades. Our large customers have used INFO for several years, and now they are catching up.

Regarding smart phones and PCs, how will the shipments look in the next few years? With 5nm and 3nm process capacity under pressure, does the company have enough capacity to support upgrades? What is the impact of AI on silicon content in the future?

Regarding the silicon content, all customers now hope to add AI on the end-side to increase the die size. The magnitude varies for different customers, but as a whole, a 10% growth is more common. We haven't seen sudden quantity growth, but we expect that AI functions will stimulate the shortening of the upgrade cycle, which may be seen in 2 years, on end-side devices such as PCs and mobile phones. As for capacity, it is indeed very tight. We are working hard to expand production support from now to 2026.

When the demand was high in 21 years and the clients were very aggressive with forecasts, how did the company cope with demand fluctuations? How did the company plan production capacity?

We have a disciplined planning process. We see that many of our customers have high forecasts, and we also use AI and machine learning to improve productivity, which is really effective. We also need our customers' products. We believe that AI demand is more real now and needed in various industries. Even so, we have checks from top to bottom to make the forecast more realistic.

Regarding the SPR plan, achieving good power consumption at the chip level, what impact does it have on system level efficiency? The more customers buy, the more they save money?

The more TSMC wafers are bought, the more money is saved. A 20% reduction in chip-level power consumption may not be reflected in a 20% reduction in system-level power consumption. The processor is the main power consumption part of the system, which is still a large part even if the system is not reduced by 20%. Therefore, customers hope to use advanced processes and migrate to 2nm.

What is the biggest bottleneck for A16 process expansion?

When we expand, we need land, electricity, and talent.

During COMPUTEX, many companies said they will accelerate the launch of their products. What is the impact on the company?

We like this trend. We are very advanced in the advanced process. Each product design takes 1.5 to 2 years, and we have long known about this. Our customers are happy, and we are also happy. We are ready and have communicated with them early to prepare for such changes.

Regarding AI chips, as die size increases, will fan out panel level packaging be a long-term layout for the company?

Indeed, we are exploring panel-level packaging, but not yet. We believe it will take about three years. Within three years, we won't have anything solid with reticle sizes ten times larger. It is currently five to six times larger. Three years ago, we introduced panel level packaging, and we will be ready for it.

Editor/Lambor