Liquid Cooling

Liquid cooling is a class of data centre cooling technologies that transfer heat from computer components using liquids rather than air, adopted at scale in the mid-2020s because AI accelerator racks exceed the cooling limits of conventional air conditioning.^[1][2] The global data centre liquid cooling market was estimated at US$2.25 billion in 2021 and projected to grow to roughly US$31 billion by 2032, reflecting the cooling demands of CPU and GPU clusters used for AI and high-performance computing.^[1]

History

Traditional data centres relied on computer room air conditioning and chilled-water systems to cool racks dissipating a few kilowatts each.^[2] As AI training and inference clusters pushed rack power densities above 30–50 kW, air cooling became inefficient, and operators turned to liquid-based methods.^[1][2] By 2025, single-phase direct-to-chip liquid cooling had emerged as a leading approach, with immersion cooling competing for hyperscale deployments.^[1][3]

Major cloud providers began fleet deployments in 2025: Microsoft started fleet deployment of direct-to-chip cooling and tested microfluidics inside Azure campuses in July 2025, and Schneider Electric introduced prefabricated modular data centre pods with liquid cooling for AI and HPC environments in November 2025.^[4]

Key Concepts

Liquid cooling systems fall into several categories:

• Direct-to-chip (cold plate) cooling: coolant circulates through metal plates mounted directly on CPUs and GPUs, removing heat at the component using conductive plates and microchannels; this is currently the most widely favoured method for AI data centres.^[1][2]
• Immersion cooling: servers are submerged in dielectric fluid, either single-phase (fluid stays liquid) or two-phase (fluid boils and condenses), allowing very high density but raising concerns about fluid handling and cost.^[1][3]
• Cooling distribution units (CDUs): manage and circulate the coolant between the facility loop and the IT loop, and are a key component of large liquid-cooled deployments.^[4]
• Water consumption: evaporative and water-based cooling can consume roughly four million litres per day for a 100 MW facility, making water usage effectiveness a central design and regulatory concern in water-stressed regions.^[5]

The shift to liquid cooling is driven by power density: racks consuming 30–50 kW or more cannot be cooled effectively with air, and liquid methods both remove heat more efficiently and reduce the energy spent on cooling.^[1][2]

Applications and Impact

Liquid cooling is used in AI training clusters, GPU-as-a-service platforms, high-performance computing centres, and increasingly in multi-tenant colocation facilities.^[1][3] Its impact extends beyond temperature control: by allowing denser racks, it reduces the floor space needed per unit of compute, and by removing the need for compressor-based air conditioning it can lower total facility energy use. Microsoft reported up to 15% energy savings in two-phase immersion trials in Azure.^[4] Analysts project the liquid cooling market will grow at roughly 25–31% per year through the early 2030s.^[1][2]

>See Also

• AI Data Centres Malaysia
• GPU Cluster
• Green AI
• NVIDIA Blackwell
• Tensor Processing Unit

References