Immersion cooling

Lore & Background

Immersion cooling reduces energy consumption by eliminating air-cooling infrastructure such as server fans, computer room air conditioners (CRACs), compressors, ductwork, and dehumidifiers, replacing them with liquid circulation pumps and heat exchangers or dry coolers. The dielectric liquids used fall into three categories: synthetic hydrocarbons, esters (natural and synthetic), and fluorochemicals. Single-phase immersion circulates the liquid without phase change, while two-phase immersion uses a low-boiling-point fluorocarbon that vaporizes and condenses on a cooling plate, transferring heat more effectively. Servers and IT hardware do not require fans in immersion; onboard fans are removed prior to submersion. Thermal pastes on heat spreaders may need replacement with solder, indium foil, or thermally conductive epoxies to avoid degradation in the dielectric fluid. A hoisting device is essential for placing and removing vertically positioned IT equipment. Until around 2014, immersion cooling was used mainly in specialized supercomputing projects such as Cray systems. The rise of high-density CPUs and GPUs for artificial intelligence, machine learning, and data mining has driven broader adoption in data centers and cryptocurrency mining.

Reader's Guide

Immersion cooling is significant for its ability to address the technical limitations of air-cooled platforms, particularly as processors and GPUs with higher thermal design power (TDP) become common. By submerging electronics in dielectric liquid, data centers can cool high-density racks that would be impractical with air, while also reducing the mechanical footprint of cooling infrastructure. The technology enables the use of entirely passive dry coolers or more efficient evaporative or adiabatic cooling towers, because the operating temperature range (up to 65 °C for servers, 75 °C for crypto miners) allows heat rejection without chillers. Power usage effectiveness (PUE) comparisons between air-cooled and immersion-cooled facilities are complicated by differing definitions: air-cooled PUE includes server fan power, while immersion PUE excludes it because fans are removed. This discrepancy can make air-cooled PUE appear overstated relative to immersion. Commercial applications span data centers, high-performance computing clusters, cryptocurrency mining, cloud hosting, and electric vehicle battery and drivetrain cooling. Immersion is also used in deep-sea research, where dielectric fluid protects electronics from corrosion and equalizes pressure in remotely operated vehicles. Major brands such as Intel and Meta have validated the advantages of submerging servers.

Did You Know?

• Immersion cooling eliminates onboard server fans, which are removed before submersion.
• Two-phase immersion uses a dielectric liquid with a low boiling point that vaporizes and condenses on a cooling plate.
• The operating temperature range for immersion-cooled servers is typically 15–65 °C.
• Dielectric liquids used include synthetic hydrocarbons, esters, and fluorochemicals.