LiquidRack: The Low-Friction Liquid Cooling Retrofit for High Density AI Data Centers

While high-density workloads demand a shift away from legacy air cooling, server-level precision spray cooling provides a scalable, low-friction infrastructure alternative to full immersion and complex cold plates.

There’s a good reason people flock to the coast in summer: water is an excellent way to cool down.

But you can’t stay at the beach all day. Eventually, most of us have to head back to a town or city. Unfortunately, cities are getting hotter. According to a recent study from Imperial College, summer heat deaths in 854 European cities have more than tripled due to climate change.

One extreme response to rising urban temperatures — aside from more ubiquitous air conditioning — would be to build municipal swimming pools on every block and encourage people to take a dip during the hottest hours. It’s a nice idea, but obviously completely impractical. Pools demand major infrastructure and, more importantly, a big behavioural shift.

Some of the same is true of immersion cooling in data centers. From a physics perspective, it’s the purest form of liquid cooling: every component is directly submerged in a dielectric fluid. But like building pools everywhere, it’s expensive, infrastructure‑heavy, and requires operators to change how they work. Other limitations around immersion include:

• Routine maintenance is harder as servers need to be lifted out of the coolant which can require specialized infrastructure
• Retrofitting immersion to existing sites can be costly and complex due to issues such as tank installation and floor loading considerations
•  Lack of standards around immersion fluids and related components

A more practical urban cooling strategy is handing out cold drinks or building out water fountains. It’s fairly simple, effective, and doesn’t require anyone to rethink their daily routine. In the data center, that’s the equivalent of cold plates: targeted cooling applied directly to CPUs, GPUs, and other hot components. Cold plate-based systems are easier to deploy than immersion but, like cold drinks, they usually don’t cool as effectively and require complex facility-water loops and external piping.

So what if there were a third option, something more effective than cold drinks but far easier to deploy than swimming pools?

Server Level Spray Liquid Cooling: Precision Thermal Management for AI retrofits

Getting the fire brigade to turn their hoses on the citizenry would be one way but a less dramatic approach is to deploy urban misting systems. These systems spray ultra‑fine droplets into the air, creating localized micro‑climates that can drop temperatures within minutes. The global misting market is projected to reach $1.3 billion in 2026 and $2.6 billion by 2033, driven by the need to cool hotter cities while conserving water. They’re now increasingly common in public spaces, hospitality, and even residential gardens. The Austrian capital of Vienna has deployed the technology as part of its ‘cool streets’ initiative.

In the data center, the equivalent is server-level precision spray cooling.

Instead of giving servers a cold drink (cold plates) or dunking them in a pool (immersion), spray cooling applies coolant directly to the surfaces that need it most; a targeted, efficient middle path. It represents the ultimate hybrid cooling retrofit strategy — allowing operators to instantly support high-density AI clusters inside their existing legacy spaces without a ground-up facility overhaul.

Spray cooling — pioneered by Airsys as part of its LiquidRack system — uses 3D‑printed precision nozzles to deliver controlled coolant directly onto CPU and GPU surfaces. This approach cuts dielectric fluid use by up to 80% compared to immersion, dramatically reducing cost, weight, and operational complexity.

This makes it a highly effective alternative for legacy air-cooled data centers where operators want to deploy high-density AI workloads but face strict budget or space constraints for secondary infrastructure. Airsys’ Tony Fischels recently described the technology: “You can think of it as you are spraying but there is a waterfall coming down the rest of the server,” he said.

Fischels went on to explain that, just like misting systems being deployed in cities, LiquidRack, a form of server-level liquid cooling, is also designed for seamless retrofits in urban and edge data centers. Crucially, it removes the need for building-wide water loops and other facility-level complexities associated with other forms of liquid cooling:

“We are really targeting the legacy data centers where they may not have enough room for adding all of that additional infrastructure on the cooling side. Also we are looking at edge deployments and enterprise facilities as well,” he said.

Obviously there is a connection between increasing global temperatures and AI: data center expansion needs to happen in as sustainable a way as possible to reduce the sector’s contribution to climate change. Highly efficient, low-PUE cooling — via technologies such as LiquidRack — is a critical part of that process.