What is liquid cooling and why are AI data centers adopting it?

Using Liquid Instead of Air to Carry Away Heat

Liquid cooling is a method of removing heat from computer hardware using fluid rather than relying solely on air moved by fans. The underlying advantage is a basic physical property: liquids can absorb and carry away heat far more efficiently than air can, for a given volume. This means liquid cooling systems can often remove heat faster and more effectively than air-based systems, particularly when dealing with components that generate a great deal of heat in a small physical space.

There are a few different approaches to liquid cooling used in data centers. Direct-to-chip cooling circulates liquid through small tubes or cold plates placed directly on or very near the hottest components, like a GPU, carrying heat away without the liquid ever touching the electronics directly. Immersion cooling goes further, submerging entire servers or components in a specially designed, non-conductive fluid that can safely be in direct contact with electronic parts while still efficiently absorbing heat.

Why AI Hardware Is Pushing This Adoption

Traditional air cooling, which relies on fans moving air across components to carry heat away, has served general-purpose data centers reasonably well for years. But AI hardware presents a more demanding challenge: GPUs and other AI accelerators generate very large amounts of heat, often concentrated densely within a single server rack, and traditional air cooling can struggle to remove that much heat quickly enough, especially as data centers try to pack more computing power into the same physical footprint.

Liquid cooling offers a way to keep up with this increased heat load without requiring vastly larger, more spread-out facilities just to give air cooling systems enough room to work. This makes it an increasingly attractive option as AI hardware density and power consumption continue to grow.

Efficiency Benefits Beyond Just Keeping Hardware Cool

Liquid cooling isn’t purely about preventing overheating; it can also improve overall energy efficiency in some implementations. Because liquid transfers heat more effectively than air, cooling systems built around liquid can sometimes achieve comparable or better cooling results while consuming less energy on things like fans and air conditioning, which matters given how much of a data center’s total energy use is dedicated to cooling rather than computing itself.

Bottom Line

Liquid cooling uses fluid instead of air to remove heat from computer hardware more efficiently, and AI data centers are increasingly adopting it because AI chips generate concentrated heat loads that traditional air cooling can struggle to manage at scale. As AI hardware continues to grow more power-dense, liquid cooling is becoming a more central part of how data centers are designed rather than a niche or experimental approach.

Go deeper

• What role does AI play in air traffic control?