CITIC Securities Co., Ltd.: Stargate (STARGATE) releases global computing power high growth

According to the Zhitong Finance APP, CITIC Securities has released a research report stating that on January 21, 2025, U.S. President Trump announced the establishment of the "STARGATE" project by OpenAI, SoftBank, and Oracle, indicating a continued boom in global AI infrastructure investment. In the short term, large-scale capital investments represented by Stargate are expected to provide strong support for the overseas computing power industry chain; in the medium to long term, domestic computing power demand is also expected to continue to increase. It is recommended to pay attention to: servers; U.S. semiconductor stocks; domestic computing power chip design and manufacturing; optical modules; PCBs; AIDC; temperature control.

CITIC Securities' main viewpoints are as follows:

The establishment of "STARGATE" will lead to a $500 billion investment in computing power over the next four years.

On January 21, 2025, U.S. President Trump announced the establishment of the "STARGATE" project by OpenAI, SoftBank, and Oracle, with an expected investment of at least $500 billion in artificial intelligence infrastructure, which is expected to create 100,000 jobs. Arm Holdings, Microsoft, NVIDIA, Oracle, and OpenAI will be the initial technology partners. OpenAI stated that the "STARGATE" project will build new artificial intelligence infrastructure for OpenAI in the U.S. SoftBank CEO Masayoshi Son will serve as chairman and stated that $100 billion will be immediately invested in AI in the U.S. The bank commented as follows:

Overseas computing power investment exceeds expectations, with Stargate's annualized investment scale comparable to North American cloud giants, while North American cloud companies' CAPEX is expected to continue to rise.

According to OpenAI's official website, Stargate plans to invest $500 billion over the next four years, with an average annual investment of $125 billion. This scale of computing power investment exceeds market expectations and is expected to further boost the development of the upstream and downstream of the industry chain. According to Bloomberg's consensus expectations, the capital expenditures of the four major North American cloud providers are close to $100 billion each in 2024, and some providers may exceed $100 billion in 2025.

From the perspective of scale, Stargate's investment in AI infrastructure is on par with that of major North American companies, which will bring significant incremental contributions to the industry. The related investments are expected to further solidify the fundamentals of the computing power industry chain development, providing a relatively solid support for the future development of the AI computing power industry chain.

The bank estimates that every $100 billion investment can build 15,000 EFLOPS of computing power: Assuming an investment of $100 billion, considering that Stargate's planned lifecycle lasts more than four years, it can be calculated based on the price of $30,000 for the latter half of the B200 chip lifecycle. Based on this price, $100 billion can purchase 3.33 million B200 chips, each with an FP16 sparse computing power of 4,500 TFLOPS (4.5 PFLOPS, where sparse computing power is twice that of dense computing power), totaling an FP16 sparse computing power of 15,000 EFLOPS. According to Tesla's earnings call, its Dojo supercomputer is expected to reach 100 EFLOPS by October 2024 The trend of domestic computing power growth is also clear, with AI applications and computing power expected to promote each other by 2025

With the rapid development of AI applications, the demand for computing power in China has also increased, and domestic investment in computing power is expected to further increase. According to the bank's calculations, the demand for AI computing power chips in China is expected to exceed 300 billion yuan by 2026.

High investment on the demand side boosts the development of the industrial chain, with core profits expected in AI computing power-related sectors

Servers: According to the description of the Stargate project on OpenAI's official website, the bank expects that more than half of the $500 billion in future capital expenditures will be used for server procurement. As initial technology partners, leading companies will play an important role in this. The bank anticipates that server manufacturers have a high market share in past supplies and are expected to further benefit from technological advantages and good past cooperation relationships;

U.S. Semiconductor Stocks: 1) Industry operating cycle: The bank expects the global semiconductor and hardware fundamentals to begin recovering in Q2 2025, with GenAI as the core driving force, and industry opportunities are expected to continue spreading around leading companies. 2) Data centers: Looking ahead to 2025, the AI industrial chain is expected to maintain high prosperity, and industry opportunities will further spread around NVIDIA, focusing on opportunities in ASIC chips, AI Ethernet, high-speed interface chips, and other sectors;

Domestic Computing Power Chip Design and Manufacturing: Computing power chips are the core components of servers. The bank expects that the prosperity of computing power demand both domestically and internationally will continue, and domestic computing power chips are expected to benefit continuously. Currently, domestic chip manufacturers are accelerating product iteration, and the cost-performance ratio of products is gradually improving. The bank expects domestic computing power chips to become more competitive in the market and gain development opportunities against the backdrop of sustained growth in AI demand; in the manufacturing sector, the domestic substitution of AI computing power chips is expected to accelerate, with advanced manufacturing's status as a core strategic asset being unprecedentedly strengthened.

Optical Modules: The establishment of Stargate clarifies the trend of long-term large-scale investment in AI computing power infrastructure in North America, which is expected to drive sustained strong demand for optical modules and accelerate the iteration and upgrading of new products. Currently, leading domestic manufacturers occupy a leading share among North American tech giants and can fully enjoy the industry benefits brought by the rise in AI capital expenditures in North America. At the same time, the accelerated product upgrades are also expected to promote the continuous improvement of the company's profitability, with related companies likely to welcome a favorable situation of rising volume and price;

PCBs: As a key component, PCBs are expected to benefit fully from AI servers, data interaction, and other dimensions, ushering in growth opportunities with both volume and price increases. From the manufacturer's perspective, high-layer, HDI, and other PCB products are currently among the core sectors that benefit most directly and significantly from AI. Companies with leading technological capacity and actively adapting to bind major customers are expected to benefit first. Specifically, domestic manufacturers that have stabilized supply are expected to continue delivering performance, and under the trend of technological upgrades, HDI manufacturers will welcome development opportunities;

AIDC: Data centers are the foundation of AI clusters, and AIDC is evolving towards park-level, high-power capabilities, with companies that have core resource reserves and bind quality customers showing significant performance elasticity. The acceleration of overseas AIDC construction is expected to promote domestic internet giants to increase investment in AI infrastructure; Temperature Control: In 2025, with the volume production of B series and domestically produced high-end chips, the penetration rate of liquid cooling will rapidly increase, driving a significant rise in temperature control value. Domestic leading companies in specific segments are expected to benefit from AI capital expenditures both overseas and domestically, with a positive outlook for enterprises that possess full-chain capabilities and advantages in core components.

Risk Factors: Risks related to the supply chain of computing power chips; risks of insufficient chip production capacity; risks of internet technology giants' capital expenditures falling short of expectations; risks of related industrial policies not meeting expectations; risks of AI application development not meeting expectations; risks of chip technology iteration not meeting expectations; risks of intensified competition among domestic GPU manufacturers, etc