CITIC SEC: US chip sanction upgrade driving the development of domestic computing power and advanced manufacturing processes.

CITIC SEC released a research report stating that the Biden administration has officially announced new global control measures for the export of AI chips. The United States has divided countries and regions worldwide into three categories and imposed different restrictions, implementing strict controls on the purchase of AI chips from mainland China. The US restrictions are driving domestic alternatives for AI computing power and advanced processes. CITIC SEC estimates that the market space for domestic AI computing chips is approximately $30 billion, and continues to recommend the related domestic semiconductor industry chain such as wafer foundries, AI chip design, semiconductor equipment, and advanced packaging. The main points of CITIC SEC are as follows: On January 13, the Biden administration officially announced new global control measures for the export of AI chips to ensure that advanced AI technology is under the control of the United States and its allies. On the evening of January 13, 2025, Beijing time, the US Department of Commerce issued a temporary final rule called the "Artificial Intelligence Diffusion Framework," which imposes new control measures on the export of advanced computing chips and closed-source AI models, with the main rules taking effect 120 days later. Restriction 1: The United States has divided countries and regions worldwide into three categories, introducing different restrictions, and implementing strict controls on the purchase of AI chips from mainland China. Specifically, this new rule classifies the destination of AI chip exports into three categories, imposes various restrictions on the quantity of AI chips that can be shipped to different countries and regions, and establishes a three-tier export restriction license system: Tier 1 includes the United States and its 18 allies (including the UK, Canada, Germany, Japan, South Korea, the Netherlands, and Taiwan), which are unrestricted and allowed to freely purchase AI chips. Tier 2 includes mainland China, Russia, North Korea, and other "Designated Restrictive" countries/regions, as well as Macau, China, which are subject to previous restrictions on AI chip purchases. Exports of high-end AI chips and closed-source AI models to these countries or regions are almost completely prohibited. Tier 3 includes approximately 120 other countries/regions (including Singapore, Israel, the UAE, etc.), which can only obtain a certain quota, limiting the maximum quantity of imported AI chips. During the period of 2025-2027, each country or region can only obtain a cumulative total of 790,000,000 TPP (Total Processing Performance, equivalent to about 50,000 NVIDIA H100 chips). Under the LPP exemption, a single buyer can order up to 26,900,000 TPP (equivalent to 1,700 NVIDIA H100 chips) per year without a license. Buyers that meet strict security standards (NVEU) may be allowed to acquire up to 5,064,000,000 TPP (approximately 320,000 NVIDIA H100 chips) between 2025-2027. Restriction 2: New export restrictions on closed-source AI models and restrictions on the global deployment of AI computing power by US data operation companies. These rules also restrict advanced closed-source AI models trained with 10^26 or more computational operations, requiring the provision of a list of end customers to obtain authorization to purchase AI chips, and strictly limiting exports and transfers to Tier 2 countries or regions. The new rules also establish a UVEU mechanism, allowing companies based in the US such as Microsoft and Google to apply for special government certification to build data centers after being authorized. As a reward for complying with certain security standards, these companies can trade AI chips more freely globally without quotas on AI chip allocation. However, to be approved, these companies must agree to retain 75% of their total AI computing power in the US or allied countries and cannot deploy more than 7% of their computing power in a single non-Tier 1 country/region. These rules aim to hinder Tier 2 countries or regions from obtaining the computing power and technology needed to produce AI from other countries, making the US and its allies the preferred location for companies to build the world's largest data centers and keeping the most advanced AI models under the jurisdiction of the US and its allies. The market space for domestic AI computing chips is $30 billion, and US restrictions are driving domestic alternatives for AI computing power and advanced processes. According to previous calculations by CITIC SEC's research department, the market size of domestic AI compute chips is approximately $20.8 billion, $30 billion, and $38.6 billion in 2024/25/26 respectively; the market size of domestic AI compute chips is approximately $7.3 billion, $16.5 billion, and $25.1 billion, with corresponding shipments of approximately 700,000, 1.15 million, and 1.57 million units. Based on the demand for domestic AI compute chips and other indicators such as chip area, we estimate that the corresponding wafer demand for domestic AI compute chips in 2025/26 is approximately 26,000/36,000 wafers. We believe that the US sanctions against AI this time generally conform to previous media reports, the market and related companies have expected it, and the overall impact is relatively limited. At the same time, we believe that the US's repeated tightening of sanctions will further drive the development of domestic AI compute chips and advanced processes, promote the progress of the domestic semiconductor industry chain, and gradually achieve a breakthrough in the advanced semiconductor industry. Risk factors: Risk of restrictions on the AI compute chip supply chain; risk of insufficient supply of advanced processing capacity; risk of underperformance in capital expenditures by internet giants; progress and strength of related industry policies falling below expectations; underperformance in the development of AI technology and applications; underperformance in chip technology iteration; underperformance in the mass production progress of domestic advanced processes; increasing industry competition, etc. Investment strategy: The content of these sanctions on AI chips is not much different from previous media reports, and the market has anticipated it. We believe that in the short term, there will be some impact on the construction of domestic AI capabilities and the progress of large models; in the medium to long term, it will help accelerate the domestic substitution of AI computing chips industry, with the core competition of AI computing power lying in advanced processes. Advanced manufacturing as a core strategic asset is strengthening its position like never before, and is expected to further accelerate the localization process of the entire advanced manufacturing industry chain. We recommend focusing on four key areas: wafer foundries, AI chip design, domestic equipment and components, and advanced packaging: 1) Wafer foundries are strengthening their position as core strategic assets for semiconductor advanced domestic substitution. 2) AI chip design companies should accelerate the layout of domestic advanced process technology and focus on domestic process layout.Faster Enterprises. 3) The trend of domestication of equipment companies is clear, and currently the most important companies to focus on are those with advanced processes, platformization, and low domestication rates in segmented markets. In terms of low domestication rates in segmented markets, companies related to measurement and testing, glue dispensing and developing, and photolithography machines should be focused on. 4) Advanced packaging plays a role in enhancing AI chip technology, with continuous technological iteration space in the 2.5D/3D/HBM related directions. It is recommended to pay attention to domestic manufacturers that are laying out advanced packaging."Hola, cmo ests?" "Hello, how are you?"