Thermal Management for Data Centers 2023-2033

With the increasing demand for high-performance computing in sectors like AI, cloud computing, and crypto mining, the thermal design power (TDP) of chips has risen significantly over the past 16 years, experiencing a four-fold increase from 2006 to 2022. In 2023, IDTechEx has observed servers with IT loads exceeding 750W and this upward trend in TDP has propelled a need for more efficient thermal management systems at both the micro (server board and chip) and macro (data center) levels. In recent years, leading data center users have collaborated with various cooling solution vendors to launch innovative pilot projects and commercialize ready-to-use cooling solutions, aiming to enhance cooling performance and meet sustainability targets by adopting more efficient cooling solutions.

 

IDTechEx’s report on Thermal Management for Data Centers covers a granular market forecast of data center cooling technologies segmented by data center power capacity. The report also provides a granular forecast of thermal interface materials (TIMs) usage by data center component.

 

Data center cooling methods can be broadly categorized into air cooling and liquid cooling, depending on the cooling medium employed. Air cooling relies on air conditioning and/or fans to blow air, utilizing convection to dissipate heat from the servers. It has been widely adopted due to its long and successful track record. However, the low specific heat of air makes it challenging to meet the increasing cooling capacity requirements. Additionally, as data center users strive to maximize rack space utilization by densely packing servers (typically 1U servers), the air gaps between servers become narrower, which further reduces the efficiency of air cooling.

 

Liquid cooling, on the other hand, takes advantage of the higher specific heat of liquid to achieve superior cooling performance. Depending on the which components the fluids contact, liquid cooling can be classified into direct-to-chip/cold plate cooling, spray cooling, and immersion cooling. Direct-to-chip cooling involves mounting a cold plate with coolant fluid directly on top of heat sources such as GPUs and chipsets, with a thermal interface material (TIM) applied in between. Cold plate cooling can achieve a partial power use effectiveness (pPUE) ranging from 1.02 to 1.20, depending on the specific configuration.

 

An emerging alternative is immersion cooling where the servers are fully submerged in coolant fluids, enabling direct contact between the heat sources and the coolant, thereby achieving the best cooling performance with the lowest pPUE of 1.01. However, its widespread adoption is still limited due to challenges such as complexity, limited industry expertise, and high upfront costs (in terms of US$/Watt). Nevertheless, immersion cooling holds potential for long-term energy savings, which is not only economically beneficial given the current context of energy crisis but also helps the large companies to achieve their energy saving and sustainability goals in the long run. IDTechEx’s comparative analysis of the total cost of ownership (TCO) between traditional air cooling and immersion cooling reveals that the payback time for immersion cooling is approximately 2.2 years, subject to the assumptions listed in the report. This report also highlights strategic collaborations and pilot projects between data center end-users and immersion cooling vendors, as well as other barriers hindering the widespread adoption of immersion cooling. Market adoption and revenue forecasts are provided for air, cold-plate, and immersion cooling through to 2033.

 

Liquid cooling can also be classified into single-phase and two-phase cooling. Two-phase cooling generally exhibits greater effectiveness, but it also presents challenges such as regulations regarding two-phase immersion cooling fluids based on perfluoroalkyl and polyfluoroalkyl substances (PFAS), mechanical strength requirements for fluid containers to withstand increased pressure during phase changes, fluid loss due to vaporization, and high maintenance costs. This report provides an analysis of different liquid cooling vendors, coolant fluid suppliers, and data center end-users, offering insights into the opportunities and threats associated with single-phase and two-phase direct-to-chip/cold plate and immersion cooling.

 

 

 

IDTechEx anticipates rapid growth in the adoption of liquid cooling, driven by factors such as the increasing power capacity of data centers, the rise of hyperscale data centers, and the availability of ready-to-use liquid cooling solutions. Specifically, cold plate cooling is expected to experience the largest growth due to its cost effectiveness and compatibility with existing air-cooled data centers, eliminating the need for extensive retrofitting to accommodate immersion cooling solutions.

In line with these projections, this report offers a detailed 10-year revenue forecast for hardware related to liquid cooling in data centers segmented by different tiers of data center IT power capacity.

 

With greater adoption of liquid cooling, new opportunities are emerging, leading to strengthened collaborations among companies involved in the data center cooling supply chain. Component suppliers such as coolant distribution units (CDUs) vendors, pump vendors, and coolant fluid suppliers are expected benefit from the increased adoption of liquid cooling. CDUs and pumps are critical components for controlling the flow rate in liquid cooling systems. Factors such as material compatibility and pressure drop need to be carefully considered. This report introduces various commercial in-rack and in-row CDUs, accompanied by a comprehensive comparison of coolant fluids based on their dynamic viscosity, density, specific heat, thermal conductivity, as well as required pipe length and pipe diameter. The report also provides a forecast of CDUs and pumps in the data center industry for the next 10 years.