AI, the strongest driving force for future growth (Taiwan Semiconductor 23Q4 conference call)
Taiwan Semiconductor (TSMC) released its fourth-quarter earnings report for 2023 (ending in December 2023) before the US stock market on the afternoon of January 18, 2024. The key points of the conference call are as follows:
1. Incremental information from the conference call of Taiwan Semiconductor (TSMC):
1) Performance outlook: The company expects revenue growth of over 20% in 2024, with CAPEX between $28 billion and $32 billion in 2024. The 2nm technology is expected to enter mass production in 2025, and the company will also develop backside power supply technology, which is more suitable for HPC. It is expected to provide to customers in 2025H2.
2) Relationship with Intel: Intel is both a customer and a competitor. The company maintains technological leadership and has a broader customer base.
3) Factors affecting gross margin: There are two negative factors affecting the gross margin. First, the production volume of N3 will increase, and the dilution of N3 in the second half of the year will be 3-4 percentage points. Second, a significant amount of N5 capacity will be converted to N3 in the second half of the year, resulting in a 1-2 percentage point impact. In the long term, a gross margin of 53%+ is achievable.
2. Full text of the Taiwan Semiconductor conference call:
2.1 Management's remarks:
Revenue situation:
Overall:
Revenue for Q4 2023 was $19.62 billion, an increase of 13.6% QoQ and a decrease of 1.5% YoY, exceeding the previous guidance of $18.8 billion to $19.6 billion. The gross margin was 53%, within the guidance range of 51.5% to 53.5%. Revenue from 3nm accounted for 15%, 5nm accounted for 35%, and 7nm accounted for 17%, totaling 67%. HPC and smartphones showed significant QoQ growth, with HPC revenue increasing by 17% and smartphone revenue increasing by 27%. For the full year of 2023, CAPEX was $30.4 billion, and it is expected to be between $28 billion and $32 billion in 2024. 70-80% will be used for advanced process capacity expansion, 10-20% for specialty processes, and the rest for advanced packaging, testing, photomasks, etc. It is expected that the revenue will grow by approximately 20-25% in 2024, with N3 revenue at least tripling and accounting for around 15% of the total revenue. The gross margin will be affected in the second half of the year, mainly due to the ramp-up of 3nm production capacity and the transfer of some 5nm capacity to 3nm. It is estimated that the revenue contribution from AI chips will reach 15%-20% or even more in 2027. There may be a second factory in Japan, and the consideration for a second factory in the United States is still ongoing, mainly depending on government subsidies.
In Q4 2023, the revenue increased by 14.4% QoQ in terms of New Taiwan Dollars and 13.6% in terms of US Dollars. The gross margin decreased by 1.3 percentage points, mainly due to the ramp-up of 3nm production. The operating profit margin decreased by 0.1 percentage points to reach 41.6%, higher than our guidance, mainly due to higher operating leverage, with an ROE of 28.1%. In terms of process technology, the revenue contribution from 3nm is 15%, while 5nm and 7nm account for 35% and 17% respectively. The advanced process technology, 7nm, has a higher proportion, reaching 67%. Looking at the whole year, the revenue contribution from 3nm accounts for 6% of wafer revenue, 5nm accounts for 33%, and 7nm accounts for 19%. The proportion of advanced process technology is 58%, compared to 53% in 2022.
Compared to the third quarter, the gross margin in the fourth quarter decreased to 53%, mainly due to the continuous ramp-up of 3nm production. Looking at the situation in 2024, the good news is that the utilization rate will increase, but N3 will dilute our gross margin, which will be around 3-4% for the whole year, but the revenue contribution will increase. Most of this will happen in 2H 2024, and our gross margin will be diluted by 2-3 percentage points in 2H 2024. Exchange rates may also have an impact. In the long term, considering our global strategy, our long-term gross margin will still be above 53%.
By platform:
HPC revenue increased by 17% QoQ, accounting for 43%. Smartphone revenue increased by 27% QoQ, reaching 43% of the total revenue. IoT revenue decreased by 29% QoQ, accounting for 5% of the revenue. Automotive revenue increased by 13% QoQ, accounting for 5%. Looking at the whole year, smartphone, IoT, and DCE revenue decreased by 8%, 17%, and 16% respectively, while HPC remained stable and automotive electronics increased by 15%. HPC accounted for 43% of the total revenue for the whole year, smartphones accounted for 38%, IoT accounted for 8%, and automotive electronics accounted for 6%. Cash and cash equivalents amounted to $55 billion, and short-term liabilities decreased mainly due to a decrease in accounts payable. Inventory turnover decreased by 11 days to 85 days, mainly due to 3nm wafers.
CAPEX:
In 2023, we spent $30.4 billion, which is lower than the previous guidance, as we continue to tighten CAPEX. The CAPEX for 2024 is expected to be between $28-32 billion. 70-80% of the CAPEX is for advanced process technology, 10-20% is for specialty processes, and 10% is for advanced packaging, testing, mask production, etc. Depreciation will increase, with a depreciation expense ratio of 30%, mainly due to 3nm. Taiwan Semiconductor's higher CAPEX and increased growth in the second year are related to our previous rapid increase in CAPEX, which has allowed us to seize opportunities in HPC AI and 5G. Despite the challenges in the market in 2023, we expect our revenue to grow at a CAGR of 15-20% in the coming years, and the increase in CAPEX in 2024 is our way of capitalizing on growth opportunities. We will continue to increase cash dividends. The dividend for 3Q23 will be distributed in April 2024, and the dividend for 2024 is expected to be NT$13.5 per share. We will maintain stable dividend growth.
Process Progress:
Currently, our most advanced process is N3, and N3E has entered mass production since 23Q4. We also have N3P and N3X, which are designed to meet the demands of HPC customers with better energy efficiency. N3P is similar to Intel's 18A process but will be released earlier, with better cost and maturity. We believe that Intel's most advanced process can only rely on product optimization, while our customers are more diverse, leading to higher process maturity. There is great interest in N2 from customers, and we expect it to enter mass production in 2025. We are also developing backside power supply technology to better serve the HPC market, and it is expected to be available to customers in 2025H2.
AI:
The global economic downturn, inflation, and the semiconductor inventory cycle have all had an impact on the semiconductor market in 2023. The semiconductor market, excluding storage, decreased by 13%, and our USD revenue declined by 8.7%. Our technological leadership has allowed us to outperform the foundry industry. We can also capture new growth markets such as AI. In 2024, we expect fabless inventory to return to normal levels, but with a weak macroeconomic environment and potential political risks, there may still be pressure on the end consumer market. We anticipate a healthier growth year in 2024, led by our leading 3nm process, strong demand for 5nm, and AI. The low base in 2023 will result in semiconductor market growth of over 10% for the full year in 2024. The foundry market is expected to grow by approximately 20%. With our advanced technology and strong customer support, we will outperform the foundry industry in terms of growth, achieving quarterly growth and an annual revenue increase of low to mid 20% in USD terms.
N3 and N3E:
N3 is currently the most advanced process, and we have collaborated with almost all HPC and smartphone customers on the 3nm process. N3 has successfully entered mass production and will reach a 6% revenue share for the full year in 2023. N3E will improve energy efficiency and enter mass production in 23Q4. The demand from smartphone and HPC customers is strong. We expect N3 to at least double its revenue share, reaching a mid-teen percentage. We will also develop N3P and N3X. The demand for our 3nm process is expected to remain strong for many years, and each of our 3nm processes will contribute to long-term revenue growth. N2:
The demand for AI in 2023 is very promising, with energy consumption and computing power being the key factors. Regardless of the approach, AI models are becoming increasingly complex, leading to a growing need for inference and training. AI models require stronger hardware support and the most advanced manufacturing processes. Our continuous technological advancements have solidified our leading position and captured the AI market. The high energy efficiency of our products has made customers rely on us, as we can provide the most advanced technology. Additionally, the development of manufacturing processes has allowed customers to engage with us at an earlier stage. Almost all AI companies are in contact with us, showing great interest in N2. Compared to N3, N2 has more interactions with HPC and intelligent mobile device customers. N2 is currently the most advanced in the industry, with the best line width and energy consumption. It is expected to enter mass production in 2025 and is currently in normal development. We have also developed dedicated solutions for HPC using N2, which will be completed in 25H2 and enter mass production in 2026. N2 will consolidate our technological leadership and enable us to seize future AI opportunities.
Distinctive Manufacturing Processes:
Currently, around 70% of our revenue comes from advanced manufacturing processes, and this proportion will increase with the mass production of N3 in the future. Mature manufacturing processes currently account for about 20% of our revenue. We collaborate with strategic customers on mature manufacturing processes to provide differentiated solutions and establish distinctive manufacturing processes. These distinctive processes will enable our mature manufacturing processes to achieve profitability on par with the company average. We anticipate that 28nm will be used for embedded storage, and we will also establish overseas factories to support local customer demands.
We will maximize shareholder interests based on customer needs and the support of local governments. In Japan, we are constructing factories for distinctive manufacturing processes, mainly for the 12nm, 16nm, 22nm, and 28nm processes. The groundbreaking ceremony is scheduled for February 2024, and mass production will proceed as planned. We have ongoing communication with the U.S. government regarding subsidies and have received substantial support for infrastructure and supply chains. We continue to collaborate with local partners, such as the Arizona Construction Association. These collaborations enhance our staff training and site safety, creating a win-win situation. The mass production of our N4 factory in 25H2 is proceeding as planned, and we expect the product quality in Arizona to be on par with Taiwan. In Europe, we are continuously constructing factories in Germany for automotive-related applications. We maintain communication with the local government, and the construction plan is proceeding as scheduled, with groundbreaking planned for Q4 2024. In Taiwan, we continue to invest in advanced manufacturing processes to support customer demands. Due to the strong demand for N3, we are expanding N3 production in Tainan. We are also preparing for N2 mass production in 2025 and planning to establish multiple N2 factories in Hsinchu, Kaohsiung, and Taichung to support strong customer demand. The Taichung factory is also proceeding as planned. We are confident in minimizing costs and creating value for shareholders.
2.2 Q&A
Q1: How will the company maintain its technological leadership in the future? How does it compare to Intel? How is the capacity planning for Intel in the future? A1: Intel is both our customer and competitor. Their 18A process is similar to Taiwan Semiconductor's N3P process, and we have also reviewed their specifications. We believe that the maturity of the processes is different. Our latest process has been mass-produced on a large scale for 25 years. I don't want to comment specifically on Intel, but we can still maintain technological leadership and have a broader customer base because almost all customers cooperate with us.
Q2: What is the reason for the company's aggressive CAPEX?
A2: Our CAPEX is $28-32 billion, which is for the capacity preparation of N2 and N3. We still believe that 18A is similar to our N3P. 18A is only used for their own products. Generally, IDM optimizes the most advanced processes on their own. We can optimize according to the customer's products. Therefore, we can optimize for AI edge, AI mobile processors, etc., and we can achieve better PPA. Intel becoming our customer also demonstrates this point.
Q3: AI is a significant technological advancement. Can competitors enter and narrow the gap with Taiwan Semiconductor?
A3: Our competitors are also using AI. Various AI companies in the United States are developing AI. AI is still in its early stages. From the end of 2022 until now, it is just the tip of the iceberg. 1nm or below is challenging, but we have new technology, which is using AI to accelerate scientific research and development. We have been laying out for many years. The development of our technology is going smoothly, and we will know the situation of 25 years in June. We are now encouraging the future use of AI in our company and also working on digital excellence. With the diligence of Taiwanese engineers and AI, we can bring semiconductor development to a new stage.
Q4: How can the company gain more value in the AI trend? Will it come more from front-end process node wafer production or advanced packaging?
A4: Revenue is contributed by both the front-end and back-end. Our customers are very successful in the AI field, and Taiwan Semiconductor is also crucial to AI. Almost everything related to AI is produced by us. As for how we capture value, the proportion of AI in data centers is increasing, and for us, the CAGR of AI is almost 50%. We are confident in seizing more opportunities. This is what we mean by achieving tens of times of revenue within a very narrow range by 2027. AI processing capabilities will be present in edge devices such as mobile phones and PCs, and there will be significant changes in chips. Previously, the CAGR was in single digits.
Q5: Regarding technological development, different companies have divergent technologies, such as transistor structures and High NA. What are the advantages and disadvantages of the company's process development? How does it compare to competitors in the future? A5: We have always made the right decisions. Technology itself does not generate value. The key is what we can provide to our customers. We offer products with the best energy efficiency and performance, and the cost is also acceptable. The maturity of the technology is also important. We consider factors such as cost and implementation of new tools like High-NA and new structures like new transistor structures. This allows us to make the right decisions at the right time to support our customers. Apart from Intel, all of our customers are satisfied with our technological development.
Q6: Why is the AI proportion estimated to be in the high-teens now, when the company previously believed it would be in the low-teens in 2027?
A6: Since last year, the demand for AI has been growing rapidly, and our customers have responded quickly, requiring sufficient capacity in both the front-end and back-end processes. We are still in the early stages of AI, and we see strong development momentum. The proportion of AI will continue to increase, possibly even more than the high-teens.
Q7: Regarding gross margin, the company's gross margin during the downturn cycle is much better than before. How do you expect the gross margin to be when operating at full capacity? How will the capacity conversion in the second half of 2024 dilute the gross margin?
A7: There are two negative factors affecting the gross margin. First, the output of N3 will increase, resulting in a dilution of 3-4 percentage points in the second half of the year. Second, a significant amount of N5 capacity will be converted to N3 in the second half of the year, which will have a 1-2 percentage point impact. The dilution of N3 will gradually decrease, and it will continue to grow in the future. The conversion from N5 to N3 is a one-time short-term impact, which will bring us long-term profitability. This is only a short-term impact on the gross margin. In the long run, our capacity construction is based on long-term market observations and cost reductions, which will enable us to achieve a high utilization rate. We believe that a long-term gross margin of 53% or higher is achievable.
Q8: The company's gross margin in 2022 is still very high. With the recovery of capacity utilization, can the gross margin return to nearly 60%?
A8: Our capacity is based on customer demand. The industry faced significant challenges last year, and everyone has gained experience in the past few years. We hope that such a possibility exists.
Q9: What is the long-term sustainability of Intel's products being manufactured by the company?
A9: We have considered all scenarios, including the possibility of Intel doing everything themselves. We have taken a conservative approach to prepare our capacity.
Q10: In terms of CAPEX, will there be a pulse-like growth after entering N2? Under the influence of geopolitical factors, will the company build a factory for N3 in Japan?
A10: CAPEX can vary greatly depending on different circumstances. In terms of capital intensity, 2021 was the peak, exceeding 50%. It then decreased to 47% and 43%, and it will be around 30% in 2023. We are still in communication with the local government regarding the plans in Japan, and it may involve the 7nm and 16nm processes. Q11: Will 3nm be mass-produced in Arizona?
A11: The second factory is still under construction, and the specific process is still under discussion. It depends on the subsidies provided by the US government. The current plan is for 2027-2028. The choice of process for overseas factories depends on the needs of local customers, and there are no definitive factors yet.
Q12: Under the influence of global geopolitics, there has been a significant increase in mature process capacity. Is there an oversupply of mature processes in the industry? What is the company's strategy?
A12: It is indeed possible that there is an oversupply of capacity. The possibility of oversupply exists. The difference is that our mature process technology will expand production, unlike our competitors. We work with our customers to create efficient and appropriate production capacity. Even if there is excess capacity in the industry, there is still demand for our mature processes. Currently, our customers are performing well, so we are not particularly concerned. This is not just a potential issue for us, but for the entire industry.
Q13: The progress of the company's 3D SOIC is good. What are the plans and capacity?
A13: The demand is still strong, and our current capacity is not enough to support our customers. We are working hard to expand production. Even if our capacity doubles, it is still not enough, so we will continue to expand production next year. We have been engaged in advanced packaging for nearly ten years, and we plan to achieve at least a 20% CAGR in the capacity of cowos 3dic soic in the next few years. We are confident that our capacity can support our customers.
Q14: What are the plans for COWOS next year?
A14: We will discuss that next year.
Q15: The company's N3 technology is excellent. Will it increase the proportion of Intel's foundry business?
A15: We have considered all possibilities, and we will also consider expansion in our production capacity.
Q16: Regarding advanced packaging, is COWOS-S currently the mainstream? Are there any changes in gross margin with the use of R/L?
A16: We are working closely with our customers, and we are also developing the next generation. Although our capacity is not fully met, we are making progress.
Q17: What is the outlook for the full-year gross margin?
A17: We have already mentioned the gross margin, and the dilution of gross margin will be greater in the second half of the year. We cannot provide a full-year outlook at this time. Our long-term plan is still 53%, and we are confident in achieving it.
Q18: What downstream market is driving the company's revenue growth this year?
A18: HPC is the largest growth driver, with a higher growth rate than the overall company. Other growth rates are slower than the overall company.
Q19: Are the customers in Arizona only in the United States and require wafer production in the United States?
A19: The factory in Arizona is prepared for all customers, but the main customers are in the United States.
Risk Disclosure and Statement for this article: Dolphin Research Disclaimer and General Disclosure