Waterloo’s Boiling Drops Could Cool AI Chip Hotspots Fast

Every AI query warms a chip somewhere, and the chips that run them are getting hotter. A University of Waterloo team now says a single drop of boiling coolant, placed exactly where a processor spikes, could keep data centre hardware cooler without paying for expensive fluid across the whole tank.

On November 10, 2026, the university released results of a hybrid immersion cooling concept from its engineering faculty, published in the Journal of Colloid and Interface Science. In lab tests, one drop reduced the temperature of a simulated chip hotspot by up to 17 degrees Celsius in 15 seconds, several times faster than conventional immersion, according to the release.

The problem: hotspots that move faster than coolant flows

Operators of AI data centres increasingly submerge chips and processors in specialized liquids to carry heat away. The Waterloo release points out a catch. Depending on the workload, small areas of a chip can become much hotter than the rest within seconds. When that happens the chip can slow down, age faster and become less reliable, and coolant that simply flows past is too slow to react.

That gap is what the team set out to close. Rather than redesign the whole cooling loop, the researchers asked whether the expensive, highly effective part of cooling could be used only when and where a spike occurs.

How the hybrid system works

The concept splits the job in two. General immersion cooling uses inexpensive deionized water to handle the steady, everyday heat. Targeted spot cooling uses drops of a costly electronics coolant to deal with sudden spikes. “It’s a division of labour,” said Dr. Sushanta Mitra, a professor of mechanical and mechatronics engineering at Waterloo. “Water is excellent at handling steady, everyday heat and the drops take care of sudden spikes.”

The team, which also included postdoctoral researcher Dr. Tian-Yu Zhang and PhD student Arnav Banerjee, built on earlier work to encapsulate one liquid in a thin skin of another. They chose a coolant with a much lower boiling point than water. It is also heavier than water, so it does not mix and stays put when placed on a hotspot.

What the lab tests showed

In laboratory testing, a single coolant drop boiled on a simulated chip hotspot submerged in a tank of water. The hotspot dropped by as much as 17 degrees Celsius within 15 seconds. Mitra explained the mechanism: boiling soaks up a lot of heat, and the rising bubbles stir the surrounding water so it carries heat away faster. “We get the punch of boiling, but only where and when it’s needed,” he said.

The researchers then wrapped the coolant drops in a thin skin of silicone oil using their encapsulation technique. Adding tiny, heat-conducting ceramic particles to that coating helps control how quickly the drops boil. These are laboratory results on simulated hotspots, and the release describes the system as under development rather than ready for deployment.

Two ways it could be deployed

The team envisions the system ultimately working in one of two ways. In the first, coolant drops circulate through the water and activate when they encounter hotspots. In the second, drops are delivered directly to hotspots that sensors have detected. Either way, the costly fluid is used sparingly while water does most of the work.

The university adds that the hybrid approach could also apply beyond data centres, including to electric vehicles, batteries, high-powered lasers and LED lighting, all fields where heat can build in small areas quickly.

What it means

For data centre operators, the appeal is economic as much as thermal: reserve premium coolant for the moments that need it and lean on cheap water the rest of the time. For chip designers and hardware teams, faster hotspot control could support steadier performance and longer component life, the two risks the release names. For energy and sustainability planners, better cooling efficiency matters because cooling is a major part of what it takes to run AI infrastructure.

Caution is warranted. The release reports lab tests on simulated chips, and it does not give cost figures, long-term reliability data, or a timeline for pilots. Questions such as how the drops are produced at scale, how they are tracked or delivered, and how the coolant is recovered will matter before any real facility adopts the idea.

The bottom line

Waterloo’s concept treats cooling like triage: use water for the everyday load and a precisely placed boiling drop for the emergency. If it holds up outside the lab, it could make cooling dense AI hardware cheaper and faster to respond. For now it is a promising, peer-reviewed step that is still early in its journey.