Q: What are the core updates in the structure of the GB200 PCB sector?
A: The structure of the GB200 PCB sector brings some significant updates that have a notable impact on the entire PCB industry. First, the usage of PCBs is expected to significantly increase in the GB200 structure.
In a rack configuration with 72 GPUs, each rack contains multiple basic components, with the computertree structure being the core part, where each tree includes two PCB units, each equipped with two GPUs and one CPU.
Additionally, NVIDIA’s switchtree serves as an intermediary for information exchange between GPUs, and InfiniBand (IB) switches are used for network connections, all of which are key areas for PCB applications.
Q: Will copper wires replace PCBs, and what is the price-volume relationship between the two?
A: While experts in the market discuss the possibility of copper wires replacing PCBs, we believe this view is overly simplistic. In the GB200 structure, although copper wires are introduced, ASIC chips must use PCBs because the line width and spacing of copper wires cannot meet PCB standards.
As the integration of electronic products increases, the role of PCBs is irreplaceable. Furthermore, from an economic perspective, the value of PCBs is expected to increase significantly, especially in terms of the increased bandwidth of switch ASICs.
Q: How is the value of PCBs calculated in the GB200 structure?
A: In the GB200 structure, we determine the corresponding PCB specifications and value by analyzing the bandwidth on each board. For example, the PCB bandwidth carrying GPUs and CPUs is about 900GB per second, requiring high-density interconnect (HDI) technology. The bandwidth of switch PCBs is approximately 57.6TB, leading to a calculation that one rack contains two IB switches.
The main PCB components of a rack include 36 PCB units, each with a bandwidth of 0.9TB, priced at about 20,000 RMB per square meter, with a total value of 72,000 RMB. The unit price of the NVLink switch PCB is about 60,000 RMB per square meter, with a value exceeding 100,000 RMB. The total value of two IB switches is 180,000 RMB.
Q: How does the value of PCBs change in larger-scale clusters?
A: In larger-scale clusters, such as configurations connecting 8 racks, the overall PCB value can reach over 3 million RMB, with each GPU corresponding to a PCB value of about 5,300 RMB. This is nearly a 300% increase compared to a single rack configuration, mainly due to the increase in L2 layers and more levels of IB switch networks. This change indicates that as cluster scales expand, the value of PCBs also increases.
Q: What factors mainly influence the increase in PCB value in the GB200 structure?
A: The increase in PCB value in the GB200 structure is mainly influenced by the increased bandwidth of switch ASICs. As bandwidth increases, the performance requirements for PCBs also rise, driving up the value of PCBs. Additionally, the GB200 structure enhances the connection layers between GPUs, leading to an increase in the number of switch ASICs, which is also an important factor in the increase of PCB value.
Q: What is the basis for market claims that copper wires may replace PCBs?
A: Market claims that copper wires may replace PCBs are primarily based on the use of copper wires in certain applications. However, this view overlooks the irreplaceability of PCBs in high-performance computing, high-speed data transmission, and other aspects.
Rack servers have never widely adopted PCB backplanes, and NVIDIA’s GB200 structure has begun using copper wires, but this does not mean that PCBs will be completely replaced. With the increasing integration of electronic products, the importance of PCBs will be further highlighted.
Q: Has there been any change in power supply requirements for the GB200 structure?
A: Regarding the power supply issues of the GB200 structure, there are currently no obvious signs indicating that its power supply requirements have significantly changed compared to before. The design of power supply typically adjusts based on the power consumption and performance requirements of the hardware.
Due to the increase in the number and performance of GPUs in the GB200 structure, there may be higher demands on the power supply system, but the specific power supply requirements still need to be determined based on actual product specifications and designs.
Q: What is the performance of each main brand in 2023?
A: Overall, products across all price ranges have shown good growth. Responding to external environmental challenges requires local adjustments, and after adjustments, product growth has been relatively satisfactory, such as Carlsberg’s high double-digit growth and good growth of local premium brands like Lebao. The company has achieved continuous growth in sales, revenue, and profit for six consecutive years. The market grew by 0.3% in 2023, and the volume of listed companies increased by 5%, outperforming the overall industry.
Q: How do you view this year’s industry competition situation?
A: From 2018 to 2023, volumes have not returned to pre-2019 levels, but the revenue and profits of beer companies have increased significantly, indicating that premiumization is a trend in the industry, and we can learn from the development history of the foreign beer industry. Consumers’ demand for personalized beer contributes to premiumization.
Medium to long-term premiumization is an inevitable trend, but premiumization does not only refer to selling more expensive products; it is about innovation, providing different experiences to consumers through products and scenarios for added value. The company will continue to adjust to meet consumer needs.
Beer is still cheaper compared to some beverages and is a relatively popular product. The growth of 0.3% in 2023, rather than a double-digit decline, indicates that there is still resilience and demand in the category, which will not be significantly affected.
Q: How do you expect product structure under this year’s mid-high single-digit revenue growth target, especially regarding the growth targets for Uusu and 1664 outside of Xinjiang? What are the current development plans for large cities?
A: From a structural perspective, the core market is developing well and is sizable. Large cities are continuously growing every year, and compared to core market categories, they are relatively simple, mainly focusing on high-end and ultra-high-end products. Therefore, the continuous increase in sales in large cities greatly promotes ton price. Currently, products in different regions will have a focus, and the company is continuously optimizing product structure and strengthening various channels.
The “Large City 2.0” strategy does not blindly pursue an increase in the number of cities (91 in 2023, with an expected 100 this year), and these 100 cities are divided into core cities, developing cities, and cities that need to be cultivated, which helps better match expectations with output.
To make returns with limited resources, the focus of investment and branding will also differ, and we expect the overall performance in large cities to improve, with higher sales further promoting premiumization.
Q: How is the ton cost performance this year? What savings in transportation costs does the Foshan factory bring? What is the expected level of cost input and how to assess the cost input-output ratio in 2024?
A: Ton cost: It is expected that raw material cost pressure will significantly improve in 2024, with barley costs down by 7%, and packaging materials will also see some decline.
Despite considering the depreciation of over 5,000 RMB brought by the Foshan factory and some costs from the acquisition of Jing A, it is still expected that ton costs in 2024 will improve. The company continues to promote cost-saving and optimization projects in the supply chain, so costs will be better controlled.
Regarding expenses: Although large cities are being rolled out in an orderly manner and expense inputs are returning to normal, overall inputs will maintain steady growth based on sales.
Q: What is the overall capacity ramp-up period for the Foshan factory and the overall capacity utilization rate in 2024?
A: The ramp-up period is about 2 months, returning to normal production levels, with a capacity utilization rate of about 77%, similar to 2023. Based on capacity utilization and supply chain network optimization, transportation costs in 2024 are expected to decrease compared to 2023.
Q: What are the capital expenditure plans for 2024 and beyond?
A: The total capital expenditure for 2024 is about 900 million RMB: 1) Mainly for the later investment in Foshan, as it was put into production in 2022, adding a capacity of about 500,000 tons. 2) Other projects, such as supply chain projects, digital transformation projects, and upgrading the ERP system.
Q: Will the high dividend strategy remain unchanged in the next 3-5 years?
A: The dividend in 2023 increased by 7.7% compared to 2022, at 2.8 RMB. The dividend situation is based on the specific situation in 2024. The company has advocated a good dividend policy over the years.
Focus on Semiconductor Equipment Catalysts
What is the main content of this conference?
The main content of this conference involves significant changes in the semiconductor sector, especially in the equipment sector, including extensive discussions on the patent applications for multi-patterning technology; the Shenyin Wanguo research team’s long-term tracking of domestic production in the photolithography machine industry chain, including research results in optical components, patent technologies, etc.; and highlighting the semiconductor equipment sector as a recommended stock this month.
What impact does multi-patterning technology have on semiconductor equipment?
Multi-patterning technology is an important catalyst in the current semiconductor equipment field, as it can greatly improve lithography precision, thereby promoting the continuous reduction of chip processes. The application of this technology may prompt semiconductor equipment companies to increase R&D investment to enhance their core competitiveness.
What is the working principle of a photolithography machine? How small of a precision can ASML’s photolithography equipment achieve?
The working principle of a photolithography machine is based on the Rayleigh criterion formula, where the CD (critical dimension) is determined by K, λ (wavelength of light), and NA (numerical aperture). DUV and EUV photolithography machines correspond to different wavelengths, with EUV lithography machines being able to achieve a drastic reduction in feature sizes due to their wavelength being only a fraction of DUV’s.
At the same time, equipment manufacturers like ASML further enhance the precision of photolithography machines by increasing the NA value and path structure design. According to ASML’s progress and roadmap, the NA of its DNA lithography machine can reach up to 1.35, and with specific configurations, the theoretical process precision can reach about 40 nanometers.
For chip processes requiring 28 nanometers, 14 nanometers, or even 7 nanometers, it is necessary to combine with the processes of wafer fabs, where the high NA version of the EUV photolithography machine has already been delivered, achieving a theoretical precision of 8 nanometers.
Why are photolithography machines the mainstream equipment for 14 nanometers and 28 nanometers?
In semiconductor manufacturing, the reason why photolithography machines are the mainstream equipment for 14 nanometers and 28 nanometers is that their basic principle involves multi-patterning. The photolithography process includes first laying a layer of photoresist on the etched medium layer, exposing it through a mask to create gaps or ablation where the photoresist is illuminated.
Next, etching occurs in the areas where the photoresist has been ablated, using the directionality of plasma to engrave the desired patterns on the underlying layer. The multi-patterning process adds an additional hard mask layer on top of the basic photolithography process, using the initial pattern of that layer as the photoresist for a second exposure and etching, thereby achieving higher density pattern transfers.
What are the differences between the LELE and FADP dual exposure processes?
The LELE (Layered Exposure Lithography) and FADP (Full-Additive Dual-Projection) dual exposure processes both achieve higher density pattern transfers through sacrificial layers or shadow masks, but their logic and principles are slightly different.
LELE requires four stages of two exposures and two etchings to complete pattern transfer, while FADP only requires one exposure, forming an initial pattern on the sacrificial layer through the sacrificial layer, then using deposition technology to add a layer of material on the existing pattern of the sacrificial layer, and finally removing part of the material of the absorbing layer to retain the wall-like pattern, which is ultimately transferred as the final photoresist pattern to the underlying medium layer.
How does the SAQP process achieve quadruple exposure?
The SAQP (Substrate Additive Quadruple Projection) process is an extension of the SAADP (Substrate Additive Double Projection) process. By adding an additional sacrificial layer and corresponding deposition processes, it further reduces the wall spacing, achieving quadruple density pattern transfers.
Specifically, the SAQP process adds a second sacrificial layer and a second round of deposition and etching processes based on the SAADP process, forming four wall-like patterns, which are then used as photoresist patterns for multiple engravings, achieving quadruple exposure effects.
However, due to yield issues and the popularity of EUV photolithography machines, quadruple exposure processes are relatively rare in practical applications, especially for domestic situations where only DUV equipment can be used. Theoretically, through quadruple exposure combined with DNA technology, it can achieve 7 nanometers precision.
Q: What is the theme of this conference?
A: The theme of this conference is about changes in the inverter industry, especially the presentation of the first quarter report and annual expectations in the research results, and focusing on the impact of inventory turning points on sector performance and changes in the production trends of different companies in the second quarter.
Q: What is the position of the inverter industry in the market?
A: Compared to other sub-sectors of photovoltaics, the profit level of the inverter industry has been fully recognized by the market. As long as the market recognizes the arrival of the inventory turning point, it becomes relatively easy to reach a consensus on the expected growth rate of subsequent sectors.
Q: How have household storage and micro-profit companies performed this year?
A: Household storage and micro-profit companies have performed well under marginal changes, especially some companies with good expectations for April and the second quarter. The market believes that this year may be a low point for the performance of these mutual assistance companies, but over a two to three-year time frame, their growth rate may be higher than the overall photovoltaic industry.
Q: How was the production trend of inverter companies like DeYe and GoodWe in April? What is the production trend of several inverter companies in April?
A: DeYe and GoodWe’s production in April was about 40% higher than the previous month, and even higher. The inventory in the first quarter was close to the bottom, but due to the lack of significant improvement in European off-season demand, production in the first quarter did not significantly increase. Several inverter companies, such as DeYe and GoodWe, showed positive production trends in April, with a month-on-month increase expected to be around 30%, and some companies even higher. For example, Yulan is expected to have a quarter-on-quarter increase in shipments of 20% to 30% in the second quarter, with an annual target of nearly 30% growth compared to last year.
Q: What is the inventory situation and outlook for the inverter industry?
A: The inventory in the inverter industry was close to the bottom in the first quarter, and companies are expected to achieve varying degrees of growth in the second quarter. Based on the current communication regarding the annual outlook, the improvements in the third and fourth quarters are quite attractive, indicating a fundamental turning point.
Q: How does the first quarter shipment volume compare to the fourth quarter? What is the inventory situation?
A: It is estimated that the shipment volume in the first quarter may be close to that of the fourth quarter, with inventory slightly weaker than the fourth quarter, at about less than 15,000 units. Since the company operates without a cover, the pressure of cost-sharing is relatively large, and it is expected that the performance in the first quarter may exceed market expectations, and the performance is still in the recovery process of consumer liquidation.
Q: If the improvement pace of the environment in April continues, how will the quantity and price improve in the second quarter?
A: If we extrapolate based on the improvement pace of the environment in April, we believe that the quantity and price improvement in the second quarter should be very significant. This will mainly manifest in a substantial improvement in quantity and a noticeable dilution of costs after the holiday. Therefore, the second quarter is expected to be quite promising.
Q: What is the performance expectation for this year? Is it better than last year?
A: This year’s performance is not weaker than last year. We believe the increase will likely exceed last year’s approximately 800 million RMB, but specific data still needs to track the production in April and May. This month’s data shows strong performance, with WeChat showing a significant increase in volume starting in March, reaching a monthly shipment of 70,000 to 80,000 units, which will help the first quarter’s performance exceed market expectations.
Q: What is the expected performance for the first quarter? What is the quarter-on-quarter growth rate?
A: It is expected that the performance in the first quarter may exceed market expectations, with a quarter-on-quarter growth rate of over 30%, even close to 40%. A preliminary estimate for the first quarter’s performance is between 300 million to 400 million RMB, with an annual shipment growth rate expected to exceed 20%.
Q1: During the 14th Five-Year Plan period, what is the annual installed capacity of distributed wind power likely to reach?
A1: During the 14th Five-Year Plan period, if the first phase chooses 10,000 villages, with each having 20 megawatts, the total capacity is 200 gigawatts. Roughly calculated, over 10 years, the annual installed capacity will not be less than 20 gigawatts.
Q2: What is the annual installed capacity of the Thousand Villages and Ten Thousand Villages project?
A2: The annual installed capacity of the Thousand Villages and Ten Thousand Villages project is at least 8 gigawatts.
Q3: Without subsidies, what will the yield of distributed wind power drop to?
A3: Without subsidies, the yield of distributed wind power may drop by 2 to 3 percentage points, roughly around 5%.
Q4: Which regions are suitable for deploying distributed wind power considering economics?
A4: Regions suitable for deploying distributed wind power are mainly concentrated in Central Plains and Southeast, such as Henan, Shandong, Anhui, Hunan, and Hubei.
Q5: What are the differences in cost and yield between distributed wind power and centralized wind power?
A5: In provinces with subsidies, the yield of distributed wind power is slightly better than centralized, with the overall investment yield being 0.5 to 1 percentage points higher. In terms of cost, distributed wind power is relatively higher, but does not require self-built internal and external lines.
Q6: What is the absolute yield of distributed wind power in the regions with the highest subsidies?
A6: In the regions with the highest subsidies, the absolute yield of distributed wind power is about 8%.
Q7: What is the yield of distributed wind power, and will this affect investors’ willingness to invest?
A7: The yield of distributed wind power is about 5%, and the high cost may affect investors’ willingness to invest.
Q8: Is distributed wind power generally subsidized by the government across China?
A8: Distributed wind power still has subsidies in a few provinces like Jiangxi, Zhejiang, Shandong, Henan, and Beijing, while other provinces have full-grid purchase but with non-fixed electricity prices.
Q9: In provinces like Jiangsu and Zhejiang, distributed wind power adopts a fixed electricity price plus subsidy model, while other provinces are market-oriented transactions. How will future policies affect yield?
A9: Distributed wind power in provinces like Jiangsu and Zhejiang adopts a fixed electricity price plus subsidy model, while others are market-oriented transactions. It is expected that future standards for the grid connection price of the Thousand Villages and Ten Thousand Villages project will be introduced, which may impact the yield.