Liquid and air cooling solutions for AI servers

Based on DIGITIMES Research's observations, the rapid increases in high-end AI servers' computing power and energy consumption are driving leaping advances in cooling technologies and accelerating their time-to-market.

On top of that, the widespread popularity of generative AI starting in 2023 is spurring even stronger demand for higher computing power. To keep up with the demand, processors come with higher and higher thermal design power (TDP), which has exceeded the limit of traditional air cooling solutions.

DIGITIMES Research believes air-assisted liquid cooling solutions have more compelling advantages in the short term as they do not require large changes to the server design and datacenter infrastructure and their costs are significantly lower compared to pure liquid cooling solutions. These advantages will make air-assisted liquid cooling solutions a top choice for datacenter operators in the next three years.

• Introduction

• Cooling demand

• Chart 1: Major players' processors and accelerators power consumption and cooling solution demand, 2009-2030 (W)

• Chart 2: Global cooling market scale and CAGR by sector, 2022-2023 and 2030 (US$b)

• Cooling technology breakdown

• Table 1: Server mainstream cooling solution types and components composed

• L2A

• Table 2: Diagram of L2A cooling solution structure

• L2L

• Table 3: Diagram of L2L cooling solution structure

• Cost

• Chart 3: Cost structure of server cooling solutions

• Comparison of key items

• Table 4: Key item comparison of major cooling technologies

• Summary