NVIDIA Blackwell Platform and ASIC Chip Upgrades Expected to Drive Liquid Cooling Penetration Rate Over 20% by 2025 | TrendForce

According to the latest report from TrendForce, the anticipated launch of NVIDIA's Blackwell platform in Q4 2024 is set to significantly increase the penetration rate of liquid cooling solutions. This rate is expected to rise from around 10% in 2024 to over 20% by 2025. The growth is attributed to heightened global awareness of ESG (Environmental, Social, and Governance) factors and the acceleration of AI server construction by CSPs (Cloud Service Providers), facilitating a shift from air cooling to liquid cooling solutions.

NVIDIA Dominates AI Server Market

In the global AI server market, NVIDIA continues to be the primary supplier, holding nearly 90% market share in the GPU AI server sector, with AMD trailing at about 8%. TrendForce notes that the initial shipment volume of NVIDIA's Blackwell platform will be limited as the supply chain finalizes product testing and optimization of high-speed transmission and cooling designs. Given the high energy consumption of the GB200 rack solution, effective cooling solutions will be critical, promoting the adoption of liquid cooling.

Challenges in Liquid Cooling Adoption

Despite the promising outlook, the current ecosystem for servers using liquid cooling remains relatively low. ODMs (Original Design Manufacturers) will need to navigate a learning curve to address issues related to leaks and cooling efficiency. TrendForce forecasts that by 2025, the Blackwell platform will account for over 80% of high-end GPU market share, stimulating competition among power supply and cooling solution providers in the AI liquid cooling market.

CSPs Accelerate AI Server Deployment

Major cloud service providers like Google, AWS, and Microsoft are rapidly expanding their AI server infrastructure, predominantly utilizing NVIDIA GPUs and proprietary ASICs. TrendForce reveals that the NVIDIA GB200 NVL72 cabinet has a thermal design power (TDP) of approximately 140 kW, necessitating liquid cooling solutions, primarily through Liquid-to-Air (L2A) methods. For Blackwell servers like HGX and MGX, which have lower density, air cooling remains the primary solution.

Liquid Cooling Initiatives by Leading Firms

Google stands out as a frontrunner in adopting liquid cooling solutions, using both air and liquid cooling for its TPU. Suppliers like BOYD and Cooler Master provide the cold plates for Google’s systems. In mainland China, Alibaba is leading efforts to expand liquid-cooled data centers, while other cloud service providers primarily rely on air cooling for their proprietary AI ASICs.

Key Suppliers for Liquid Cooling Components

TrendForce indicates that cloud service providers are selecting key component suppliers for the GB200 cabinet’s liquid cooling solution. Major suppliers for cold plates include Qihong and Cooler Master, while manifold suppliers are Cooler Master and Shuanghong. Cooling Distribution Units (CDUs) are supplied by Vertiv and Delta. Quick Disconnect (QD) fittings, essential for preventing leaks, are currently sourced from suppliers like CPC, Parker Hannifin, Danfoss, and Staubli, with additional suppliers like Jiazhe and Fushida entering the verification phase. By the first half of 2025, these companies may join the ranks of QD suppliers, potentially alleviating current supply shortages.