NVIDIA GB200 Servers to Boost Liquid Cooling in AI Data Centers

• NVIDIA GB200 servers' high TDP drives liquid cooling from 14% to 33% penetration by 2025.
• L2A cooling transitions to efficient L2L systems starting in 2027 for better AI data center stability.
• CSPs like Google, AWS, and Microsoft expand globally with liquid-cooling-ready modular data centers.

NVIDIA is also about to relocate global usage of liquid cooling solutions in data centres worldwide, with forthcoming GB200 NVL72 rack servers.  Planned to be launched in 2025, the new top-power servers will drive the transition between experimental pilots and the scalable deployment to take on the thermal challenges of the increasingly intricate AI computation.

The surge in demand stems from the dramatic increase in power consumption for GPUs and ASIC chips. NVIDIA's GB200 and GB300 NVL72 systems boast a Thermal Design Power (TDP) of 130–140 kW per rack, far surpassing the limits of traditional air-cooling methods. This has prompted data center operators to embrace liquid cooling, with penetration rates expected to rise from 14% in 2024 to 33% in 2025, and further growth is anticipated in the years ahead.

Currently, liquid-to-air (L2A) cooling serves as the primary transitional solution, leveraging existing infrastructure while managing constraints in water circulation systems. However, the liquid-to-liquid (L2L) cooling will emerge as the dominant approach starting in 2027, offering superior efficiency and stability amid rising AI chip densities.

Also Read: L&T Tech Unveils GenAI Framework to Boost Product Development

Major Cloud Service Providers are already investing heavily. Google and AWS have introduced modular data centers equipped with liquid cooling pipelines in regions like the Netherlands, Germany, and Ireland. Microsoft is conducting pilot programs in the U.S. Midwest and Asia, with plans to standardize liquid cooling architecture from 2025 onward. These expansions span North America, Europe, and Asia, signaling a global shift towards liquid-cooling-ready facilities.

The trend is boosting demand for specialized components, including cooling modules, heat exchange systems, cold plates, Coolant Distribution Units (CDUs), and quick disconnects (QDs). Suppliers such as Cooler Master, AVC, BOYD, and Auras are ramping up cold plate production, while Delta and Vertiv lead in CDUs. Quick disconnect providers such as CPC, Parker Hannifin, Danfoss, and Staubli are key players in NVIDIA's GB200 ecosystem.

As AI workloads intensify, this evolution underscores the industry's commitment to sustainable, high-performance computing, potentially reshaping data center designs for decades to come.