AI Chip Liquid Cooling Adoption to Hit 53% by 2026; Google’s Servers Over 80% Liquid-Cooled

According to a report by TrendForce, as AI chipmakers such as NVIDIA, AMD, and Google continue iterating their AI chips, the thermal design power (TDP) per chip has commonly exceeded 1 kW, with rack-scale AI servers now reaching power levels of several hundred kilowatts. Liquid cooling is becoming the standard configuration for high-end AI infrastructure, with its adoption rate in AI chips expected to rise from approximately 33% in 2025 to 53% in 2026. Google has taken the lead among cloud service providers in adopting liquid cooling technology, with over 80% of its AI servers already utilizing liquid cooling. The company has developed a highly customized liquid cooling architecture comprising components such as cold plates, manifolds, and cooling distribution units (CDUs). NVIDIA’s VeraRubin platform fully adopts a fanless, all-liquid-cooling architecture, extending liquid cooling coverage beyond GPUs and CPUs to critical components including CX9 NICs, power distribution boards, and optical transceivers. Current major suppliers of cold plates include Cooler Master, AVC, BOYD, and Auras. AVC is expected to begin shipping VeraRubin cold plate modules in the third quarter and has already entered the AI ASIC platform supply chains of AWS, Google, Meta, and OpenAI. NVIDIA originally planned to use a two-piece vapor chamber design for its Rubin platform but temporarily shifted to a single-piece solution due to substrate warpage issues. This vapor chamber is currently supplied exclusively by Jentech.