Liquid Cooling Penetration in AI Infrastructure Projected to Reach 53% by 2026
According to a TrendForce report, liquid cooling is rapidly becoming the standard for high-end AI infrastructure, with its penetration rate in AI chips projected to rise from 33% in 2025 to 53% by 2026. This shift is driven by next-generation AI chips from Nvidia, AMD, and Google exceeding a thermal design power (TDP) of 1kW per chip. As AI workloads demand massive computing power, traditional air cooling is no longer sufficient to manage the extreme heat generated by high-density server racks. Adopting liquid cooling significantly improves Power Usage Effectiveness (PUE) and enables data centers to support next-generation hardware like Nvidia's upcoming Vera Rubin platform. Nvidia's Vera Rubin platform will adopt a fanless, full liquid cooling architecture, extending cooling to network cards, power boards, and optical transceivers. Additionally, Nvidia temporarily reverted to a one-piece vapor chamber design for the Rubin platform due to substrate warping issues, with Jentech serving as the sole supplier.
## BACKGROUND
Liquid cooling systems in data centers rely on key components such as cold plates, manifolds to distribute fluid, and Coolant Distribution Units (CDUs) to regulate temperature and flow. TDP (Thermal Design Power) represents the maximum amount of heat a chip is expected to generate under workload, which has skyrocketed with modern GPU architectures.