As the demand for artificial intelligence computing power continues to grow, liquid cooling can better meet AIDC's cooling needs, and liquid cooling is expected to be deployed at an accelerated pace in the AIDC scenario.
The Zhitong Finance App learned that Open Source Securities released a research report saying that the B100 may debut in GTC 2024 and switch from air cooling to liquid cooling. As chip power continues to rise, liquid cooling may become the first choice. The maximum power of a single NVIDIA DGX A100 server is about 6.5 KW. With the rack loading rate unchanged, the power of the single cabinet also continues to increase due to the increase in server power, approaching the 12-15KW cooling limit. Continued use of air cooling will lead to a sharp increase in demand for air conditioning between rows. The cost and performance advantages of liquid cooling solutions in high-density cooling scenarios are remarkable, and may become the best choice. As demand for artificial intelligence computing power continues to grow, the bank believes that liquid cooling can better meet AIDC's cooling needs, and liquid cooling is expected to accelerate deployment in the AIDC scenario.
Beneficial targets: Liquid-cooled equipment: Invec (002837.SZ), Science Innovation Source (300731.SZ), etc.; Liquid-cooled servers: ZTE (000063.SZ), Ziguang (000938.SZ); Liquid-cooled IDC: Baoxin Software (600845.SH), Runze Technology (300442.SZ), etc.; cooling temperature control: Shenling Environment (301018.SZ), Imikon (300249.SZ), etc.
Open source securities views are as follows:
Short-term catalysis: chip power continues to rise, AIGC drives data center development towards high density
The B100 may debut in GTC 2024, changing from air cooling to liquid cooling. As chip power continues to rise, liquid cooling may become the first choice. AI brings demand for high computing power, driving both the number of cabinets and the power of a single cabinet. The development of data centers towards high density is expected to drive the rapid development of liquid cooling. (1) The power consumption of mainstream computing chips continues to increase. The TDP of many Intel CPUs has reached 350W, and NVIDIA's H100 SXM TDP has even reached 700W, gradually approaching the upper limit of air cooling (about 800W). The B100 TDP or up to 1000W, and the TDP of domestic computing chips is also around 300W, and liquid cooling has become an optimal or necessary option; (2) Due to networking methods and application considerations, AI clusters have a high power density. AI clusters have certain requirements for computing power density. Too scattered training units are not conducive to operation development. Reducing networking distance can also reduce communication consumables costs. (3) The power of a single cabinet continues to grow, approaching the limit of air cooling and cooling. The maximum power of a single NVIDIA DGX A100 server is about 6.5 KW. With the rack loading rate unchanged, the power of the single cabinet also continues to increase due to the increase in server power, approaching the 12-15KW cooling limit. Continued use of air cooling will lead to a sharp increase in demand for air conditioning between rows. The cost and performance advantages of liquid cooling solutions in high-density cooling scenarios are remarkable, and may become the best choice. As the demand for artificial intelligence computing power continues to grow, the bank believes that liquid cooling can better meet AIDC's cooling needs, and liquid cooling is expected to be deployed at an accelerated pace in the AIDC scenario.
Liquid cooling has many advantages over air cooling, and the ecological development of liquid cooling is accelerating.
Liquid cooling technology has significant advantages, and operators help improve the liquid cooling ecosystem. Although air cooling technology is currently the most commonly used data center cooling technology, it has the defects of low density and poor cooling capacity, and is declining in scenarios with high cooling density such as AI clusters and HPC clusters. Compared with liquid cooling and air cooling technology, liquid cooling technology mainly has advantages such as (1) low energy consumption; (2) high heat dissipation; (3) low noise; (4) low TCO; (5) high space utilization; (6) low environmental requirements and easy deployment; and (7) easy recovery of waste heat. Under the strong impetus of telecom operators, the bank believes that the liquid cooling industry chain ecology is expected to develop rapidly, solve the pain points of the industry where liquid cooling product standards are not uniform and CAPEX is high, and the liquid cooling penetration rate may continue to grow.
Long-term drive: In the context of dual carbon, liquid cooling can fully meet the policy's PUE requirements.
The policy has tightened PUE requirements to guide the green development of data centers. As the scale of data centers continues to expand and the industry's electricity consumption is increasing day by day, data center energy saving and emission reduction are imminent. Guided by carbon peaks and carbon neutrality strategies, policy documents accelerate the evolution of data centers towards greening and low carbon, and promote the rapid development of liquid cooling. Local governments have increasingly strict PUE requirements for new data centers, and old high-energy data centers are facing the fate of being eliminated or converted to low-energy cooling methods such as liquid cooling. The bank believes that in the context of dual carbon, IDC greening is the general trend. Liquid cooling or cooling technology is the only way to evolve. As PUE requirements continue to be strict, the advantages of liquid cooling and cooling are gradually becoming prominent. Whether it is IDC's new construction needs or inventory transformation needs, liquid cooling is expected to become the first choice.
Risk warning: the risk of changes in national policies, increased risk of industry competition, and the development of liquid cooling falls short of expectations.