The era of liquid cooling for servers is on the horizon, but significant challenges remain. DIGITIMES Research analyst Kaylee Chiu outlines four critical hurdles liquid cooling technology must address.
Nvidia's GB200 is set to launch, with key manufacturers Foxconn and Quanta planning small-scale production by late 2024, followed by full-scale production in the first quarter of 2025.
The GB200 servers are expected to drive substantial demand for liquid cooling solutions, which, until now, have not been widely adopted.
Four challenges in liquid cooling adoption
During the DIGITIMES Tech Forum 2025, Chiu outlined the main challenges for liquid cooling systems, including leakage concerns, increased validation complexity, lack of modularity and standardization, and longer construction timelines for data centers.
Servers have traditionally relied on air cooling for heat dissipation, despite earlier efforts by companies like IBM to introduce liquid cooling designs. However, the lack of demand for chip cooling solutions led to limited adoption of these designs.
This shift occurred with the launch of Nvidia's GB200 servers, which integrate liquid cooling as a standard feature, ushering in a new era for server cooling technology.
Chiu explained that Nvidia's current Hopper microarchitecture AI servers, which use 3D VC air cooling, can handle only up to 750 watts of heat dissipation. As a result, the shift to liquid cooling is inevitable, with liquid cooling offering up to 28 times the heat dissipation capacity of traditional air-cooled servers.
The rising demand and environmental benefits of liquid cooling
Liquid cooling is gaining attention for its superior heat dissipation compared to air cooling, as well as its environmental benefits.
Chiu emphasized that air-cooled data centers currently allocate 40% of their energy consumption to cooling systems, while adopting liquid cooling can reduce this figure to just 7%, promoting sustainability and improving energy efficiency.
Liquid cooling and data center construction outlook
The trend toward liquid cooling is gaining momentum, particularly with the upcoming GB300, which will demand even greater cooling capacity.
Despite this, liquid cooling adoption in servers remains in the single digits as of 2024, and scaling up for mass adoption in 2025 will face significant challenges.
Chiu identified four major challenges for liquid cooling, starting with leakage. Recent reports of leaks in liquid cooling systems have frustrated customers. Industry experts note that two key issues remain unresolved: preventing leaks and determining responsibility when they occur.
Another challenge is the complexity of validation. Since liquid cooling technology is still in its early stages, it requires more certification processes than well-established air cooling systems.
Nvidia has begun certifying cooling components, with companies like Auras Technology recently announcing that several of their liquid cooling products have received Nvidia certification, making them recommended suppliers.
The third challenge is the lack of product standardization and modularization, complicating the procurement process for customers like ODMs and CSPs. The fourth challenge is the extended construction time needed for data centers.
Chiu noted that constructing a data center typically takes 3 to 5 years. The industry highlights that liquid cooling requires the installation of piping, chillers, and other equipment, and will only be widely adopted in new data centers.
Mike Yang, GM of Quanta Cloud Technology, recently stated that CSP customers are accelerating the development of new data centers.
