The Hot, Wet Reality of Modern GPU Compute
Keep the chips cool or watch them melt. It is that simple.
As AI workloads scale, standard air cooling has hit a hard physical wall. Data centers are packing racks with power-hungry GPUs throwing off unprecedented heat, forcing operators to pivot to liquid cooling. But high-temperature liquid cooling loops—which use mixtures of water and proprietary chemical inhibitors—bring a nasty, dirty side effect: bacterial outbreaks.
It sounds primitive. It is.
To run GPUs hotter and squeeze every drop of compute out of the hardware, facility managers often increase the water ratio in their coolants because water is an excellent heat conductor. But warm, stagnant, or slow-moving water loops are a paradise for microbes. When bacteria take hold, they form a thick, slimy biofilm. This slime clogs the micro-channels in cold plates, blocking heat transfer and causing chip temperatures to spike.
The remedy is brutal. You have to flush the entire system.
Flushing a single high-temperature liquid loop means taking the rack offline for five or six hours. In a hyperscale setup, that downtime can bleed millions of dollars in lost compute revenue. In our industry, we already talk about the massive environmental footprints of centralized clusters. Just look at how Nvidia’s New Cooling System Cuts Water Use without actually solving the underlying grid dependency. Adding a biological tax to the operational equation makes the margins even tighter.
For years, the standard way to prevent these clogs was manual sampling. A technician draws fluid, puts it in a vial, and sends it to a lab. You wait days, sometimes weeks, for a report. By the time the lab flags a bacterial bloom, the channels are already choked.
From Excavators to Data Center HVACs
Enter Omen AI. Their solution is a miniature, in-rack spectrometer that monitors coolant chemistry in real time. Instead of waiting for a lab, data center operators get constant telemetry on their fluid health.
The company's origin story is weird, but it explains the technology. Zach Laberge, the founder, is young. He started his first company at 14, raising $3 million to track heavy construction equipment before it shut down. After dropping out of high school—with the full backing of his parents, including his mother, Ontario’s former Minister of Education—he founded Omen in 2024.
He didn't start with data centers. He started with backhoes.
Laberge’s initial focus was monitoring hydraulic and engine fluid systems in construction machinery so they could flag issues before a pump blew out on a job site. If the sensor saw copper or chromium, it meant a pump was wearing down. If it saw silicon, a seal was disintegrating.
Caterpillar dealerships signed up early on. They loved the real-time feedback. But Caterpillar doesn't just sell excavators; they sell giant gas-powered generators and turbines that provide standby, on-premises power to data centers. Six months ago, those same dealers started asking Omen if the sensors could monitor the fluid loops on their building turbines and HVAC systems.
Data centers are full of liquid. Laberge saw the writing on the wall.
Liquid-to-air heat exchangers, chiller loops, and direct-to-chip cold plates are all plumbing. If you can monitor a hydraulic pump on a bulldozer, you can monitor the fluid keeping an AMD or Nvidia cluster from frying. Omen shifted its focus.
The In-Line Chemical Telemetry Land Grab
Doing this at scale requires cheap, reliable optical hardware. Ten years ago, you couldn't put a high-precision spectrometer inside a server rack without going bankrupt. Today, cheaper optical sensors and improved signal processing software have changed the economics.
The software filters the optical noise, while the hardware is cheap enough to deploy at scale. It means you can spot chemical degradation or bacterial markers immediately, preventing catastrophic shutdowns before they start.
The venture market is noticing. Omen AI recently wrapped up a $31 million Series A round led by Nava Ventures, with backing from CRV, Vanderbilt University, Mann+Hummel, Starhill Holdings, and Hard Launch Capital. High-profile executives from General Motors, Bridgestone, Johnson Controls, and TensorWave also threw in personal cash. That brings their total funding to $40 million since their 2024 launch.
They're not the only ones moving in here. The veteran water-monitoring firm Pyxis recently launched its own dedicated data center coolant monitoring tool. The market is validating the concept because the alternative is flying blind.
Clients are testing it now. TensorWave, which runs an AMD-based AI compute cloud, is using Omen's sensors in its production infrastructure. Their president, Piotr Tomasik, made the problem clear: most of the industry is flying blind on fluid variables. Given how sovereign AI demand is forcing companies to secure power and cooling capacity wherever they can find it—something highlighted by the scale of SoftBank Charts European Sovereign AI Push with €75 Billion French Data Center Expansion—knowing exactly what is happening inside your cooling pipes is no longer a luxury. It is a survival skill.
If you don't monitor the fluid, your multi-million dollar GPU investment will end up choked by slime.