Liquid Immersion Cooling Reaches Mainstream Adoption

Full immersion liquid cooling - where entire servers are submerged in tanks of dielectric coolant - has crossed the threshold from laboratory curiosity to production infrastructure. Equinix, Digital Realty, QTS, and several other major operators now offer immersion cooling as a standard deployment option, driven by AI workloads that generate thermal loads impossible to manage with air alone. The distinction between immersion cooling and the direct-to-chip (DLC) approach matters: immersion submerges the entire server, achieving PUE improvements of 0.3-0.5 and supporting rack-equivalent densities of 100+ kW, while DLC uses cold plates on individual chips, achieving PUE improvements of 0.2-0.3 at densities of 50-80 kW.

The market for immersion cooling specifically has grown from approximately $400 million in 2024 to an estimated $1.8 billion in 2026, with projections suggesting it could reach $8-10 billion by 2028. Companies leading the immersion segment include GRC (Green Revolution Cooling), which has deployed systems for the US Department of Defense and several hyperscalers; LiquidCool Solutions, which pioneered a single-phase immersion approach; and Submer, a Barcelona-based company that has attracted investment from Schneider Electric. The broader liquid cooling market (including DLC) is valued at $4.2 billion and projected to reach $32 billion by 2028 per Dell'Oro Group.

The physics make the case compelling. Dielectric coolant has roughly 1,200 times the heat capacity of air by volume, meaning it can absorb dramatically more thermal energy in a given space. In a full immersion deployment, every component of the server - CPU, GPU, memory, storage, networking - is cooled simultaneously by the surrounding fluid, eliminating hot spots that can throttle performance in air-cooled environments. For AI training workloads where GPU utilisation rates regularly exceed 90% for extended periods, this thermal uniformity translates directly to higher sustained performance.

Operators report real-world PUE reductions from 1.4-1.6 (typical air-cooled) to 1.03-1.1 with immersion cooling - approaching the theoretical minimum of 1.0. For a 100 MW facility, reducing PUE from 1.4 to 1.1 eliminates approximately 30 MW of cooling energy consumption, which at $0.065/kWh translates to roughly $14 million per year in energy savings. Over a 15-year facility life, the energy savings alone can exceed $200 million - making immersion cooling economically compelling even before considering the performance benefits.

Hot-climate markets are particularly impacted. In Phoenix, where ambient temperatures exceed 45C for months, traditional air cooling requires massive chiller plants that consume 35-45% of total facility power. Immersion cooling reduces this to under 5% because the coolant temperature remains stable regardless of outside conditions. Similar advantages apply in Dubai, Riyadh, Mumbai, and Singapore. For operators in these markets, immersion cooling is not just an efficiency improvement - it is a viability requirement for next-generation AI deployments.

The challenges are primarily operational. Immersion cooling requires fundamentally different maintenance procedures: servers must be removed from tanks and allowed to drain before any hardware work can be performed. Coolant quality must be continuously monitored to prevent contamination that could damage components. And the physical infrastructure - tanks, pumps, coolant distribution units, heat exchangers - represents a new category of mechanical systems that most data centre operations teams have limited experience with. Training and certification programmes are being developed by vendors and industry groups, but the skilled labour shortage remains a constraint on deployment velocity.