Guangda Securities: Rapid development of data centers may exacerbate copper shortages, bullish on copper prices continuing to rise

Zhitong
2024.04.05 09:13
portai
I'm PortAI, I can summarize articles.

Guangda Securities released a research report stating that the rapid development of data centers may exacerbate copper shortages, with copper prices expected to continue to rise. It is predicted that by 2023, the global demand for copper in data centers will reach 467,000 tons, accounting for 1.7% of global copper consumption. Guangda Securities believes that although the demand for copper from data centers is relatively small, the annual incremental supply is comparable to the supply-demand gap, which may exacerbate copper shortages. In addition, Nvidia recently released the GB200 super chip, significantly increasing market attention to copper connections, and the development of the AI industry will also drive the construction demand for AI data centers. It is expected that by 2026, the electricity consumption of data centers, cryptocurrencies, and AI will significantly increase, further boosting the demand for copper

According to the Zhitong Finance and Economics APP, Guotai Junan Securities released a research report stating that, according to Cushman & Wakefield, the global data center's additional installed capacity in 2023 will be 7.1GW. With 1MW data center using 65.8 tons of copper (estimated by Schneider Electric), the global data center's copper consumption in 2023 will be 467,000 tons, accounting for 1.7% of the global copper consumption in 2023. The firm previously predicted that the refined copper supply-demand gap in 2024/2025 would be 20,000 to -140,000 tons. The firm believes that although the demand for copper from data centers is relatively small, the annual incremental supply is comparable to the magnitude of the supply-demand gap, which may exacerbate the copper shortage, and is optimistic about the continued upward trend in copper prices.

Event: Recently, NVIDIA released the GB200 super chip, significantly increasing market attention to copper connections. At the same time, the rapid development of the AI industry will also drive the construction demand for AI data centers. The firm calculated the demand for copper in emerging areas such as data centers, GB200, and other electronic communications.

Guotai Junan Securities' views are as follows:

By 2026, the data center (including AI data centers), cryptocurrency, and AI electricity consumption will increase by 74% compared to 2022 (neutral assumption).

According to the IEA, the combined electricity consumption of data centers, cryptocurrencies, and AI in 2022 is approximately 460TWh, close to 2% of global electricity consumption. The IEA predicts that by 2026, under conservative, neutral, and optimistic assumptions, the electricity consumption of data centers (including AI data centers), cryptocurrencies, and AI will increase to 620TWh, 800TWh, and 1050TWh respectively, with compound annual growth rates of 8%, 15%, and 23%.

By 2026, the copper consumption of data centers (including AI data centers) will be approximately 710,000 tons, an increase of 51% compared to 2023 (neutral assumption).

According to Cushman & Wakefield, the additional installed capacity of global data centers in 2023 is 7.1GW, and a 1MW data center uses 65.8 tons of copper (estimated by Schneider Electric), resulting in a global data center copper consumption of 467,000 tons in 2023, accounting for 1.7% of global copper consumption in 2023. The firm assumes that the future growth rate of data center installed capacity will be consistent with the compound growth rate of the aforementioned electricity consumption. Therefore, under the neutral assumption, the copper consumption of data centers from 2024 to 2026 will be 536,000/616,000/707,000 tons respectively, with annual increments of 69,000/80,000/91,000 tons (only considering copper used in equipment, not the increased demand for supporting power grids due to increased electricity generation).

The copper cable used in GB200 is approximately 3,218.7 meters.

NVIDIA's recently released GB200 super chip at GTC 2024 uses a copper connection scheme, mainly due to the low cost and low energy consumption advantages of copper connections. According to NVIDIA's information, a total of about 5,000 independent copper cables (each copper cable containing 8 sets of high-speed copper wires, 1 set containing 2 round-trip wires and 1 ground wire, about 24 copper wires) are used inside the cabinet, with a cumulative cable length of nearly 2 miles (approximately 3,218.7 meters) GB200 drives demand for copper by about 0.4 million tons (assuming annual demand of 50,000 units).

GB200 uses American standard AWG26/30 wire with a diameter of 0.405/0.255mm; assuming a silver plating thickness of 3μm (typically 1-3μm for domestic copper cables) and calculating based on a copper density of 8.96g/cm3 and a cable length of 3218.7 meters, the demand for copper metal per cabinet is approximately 86.5kg. If the global annual new demand for GB200 is 50,000 units, the corresponding annual copper demand would be 4325 tons, resulting in a relatively small impact on copper demand.

Investment advice: The rapid development of data centers may exacerbate copper shortages. The bank previously predicted a refined copper supply-demand gap (supply - demand) of 2/-14 thousand tons for 2024/2025. Although the impact of data centers on copper demand is relatively small, the annual incremental supply provided is comparable to the magnitude of the supply-demand gap, which may exacerbate copper shortages. It is recommended to remain bullish on Luoyang Molybdenum (603993.SH), Zijin Mining (601899.SH), Western Mining (601168.SH), Jin Chengxin (603979.SH), and pay attention to Tongling Nonferrous Metals (000630.SZ). As copper connectors have a relatively small impact on copper demand, it is suggested to focus on Boway Alloy (601137.SH), Chujiang New Materials (002171.SZ), and other copper processing companies.

Risk warning: Risks include delays in data center construction progress and unexpected calculation assumptions