Simon Glairy is a recognized expert in the fields of insurance and Insurtech, with a specialized focus on risk management and AI-driven risk assessment. As data centers evolve to support massive AI workloads, the physical infrastructure—specifically high-end cooling systems—is becoming a critical failure point that insurers and operators alike are watching closely. In this conversation, we explore the transition from traditional maintenance to real-time chemical monitoring, the financial high-stakes of system downtime, and how technology originally meant for heavy construction is now safeguarding the heart of the AI revolution.
Data center operators often increase water ratios in cooling systems to improve heat absorption for high-performance chips, but what are the hidden risks of this chemical trade-off?
When you push GPUs to run hotter, you have to find a way to suck that heat out faster than traditional air cooling ever could. By increasing the water ratio in the cooling fluid, you get better heat absorption, but you are essentially diluting the inhibitors that keep bacterial growth at bay. This creates a perfect, warm breeding ground for outbreaks that can quickly clog the narrow, intricate flow of a high-end cooling system. It is a claustrophobic nightmare for an operator to realize their multi-million dollar rack is suffocating because of microscopic contamination they could not see until the flow stopped. Seeing that nasty, slimy buildup firsthand makes you realize how fragile these massive digital engines really are when their chemistry is out of balance.
When a system finally fails due to a bacterial outbreak or chemical clogging, what does the recovery process look like, and how does this impact the facility’s bottom line?
The recovery process is not just a quick fix; it is a grueling, multi-hour ordeal where an entire rack might be shut down for five or six hours. During that window, the financial bleeding is intense, with potential costs reaching millions of dollars in lost compute time and missed opportunities. You are forced to completely flush the system to clear the contamination, a manual and stressful task that halts the very AI processing that customers are paying a premium to access. The silence of a downed rack is a heavy, expensive weight for any data center manager to carry in an industry where uptime is the only metric that matters. It highlights why “flying blind” on fluid health is no longer a viable strategy when the stakes are this high.
How did the technological shift from monitoring heavy construction machinery provide the necessary insights to tackle the complexities of data center cooling fluids?
The transition was remarkably organic because, at the end of the day, a gas turbine or a massive generator relies on fluid health just as much as a server rack does. Early work with companies like Caterpillar revealed that the same sensors used to spot copper and chromium from wearing pumps, or silicon from failing seals, could be adapted for the HVAC and chip cooling systems in data centers. When we realized that data centers are essentially buildings full of fluid, from the building-side cooling to the direct-to-chip liquid loops, the pivot made perfect sense. It is about taking that rugged, real-world reliability from a 14-year-old founder’s first venture in construction and applying it to the pristine environment of a GPU cloud. This cross-industry validation proved that the chemistry of mechanical failure is universal, whether it is happening in a tractor in a field or a high-performance compute cluster.
With the recent $31 million Series A round and a total of $40 million raised, how is the industry reacting to the idea of replacing lab-based testing with real-time on-site analytics?
The reaction has been one of relief and rapid adoption, as evidenced by the dozen data center customers already working to integrate this real-time offering. Traditional methods involve mailing fluid samples to labs and waiting days for results, which is a reactive approach that leaves infrastructure vulnerable to sudden shifts in fluid health. By deploying a tiny spectrometer on-site, companies like TensorWave can monitor their AMD-based AI clouds with a level of precision that was previously impossible. This real-time awareness allows operators to spot bacterial growth or pump degradation before it becomes a massive, system-wide problem. Having the respect of established, large corporations in a space that usually moves slowly is a testament to how badly the industry needed to stop guessing about their cooling chemistry.
What specific advancements in hardware and software have finally made it possible to deploy high-precision chemical sensors at the scale required by modern data centers?
We have finally reached a tipping point where optical hardware has become cheap enough to play at scale, which was a major barrier for this kind of sensing in the past. However, the hardware is only half the story; the real breakthrough is in signal processing software that lets us make sense of the “noise” in a complex liquid environment. This combination of recent improvements in optical tech and signal processing allows us to see exactly what is going on chemically without needing to extract samples. It is a sophisticated dance of light and data that gives operators a clear window into the health of their infrastructure. This technological leap ensures that as data centers get more powerful and generate more heat, our ability to protect them scales right alongside the compute demand.
What is your forecast for data center infrastructure?
I forecast that we are moving toward an era of “self-aware” infrastructure where every gallon of cooling fluid is monitored with the same intensity as the data flowing through the chips. As the demand for AI compute power continues to skyrocket, the current $40 million in investment we are seeing is just the tip of the iceberg for a market that can no longer afford even a few hours of downtime. We will see a shift where real-time chemical monitoring becomes a mandatory standard for insurance and operational compliance, moving away from manual lab tests entirely. The facilities that win will be those that treat their cooling systems as a critical, data-driven variable rather than just a secondary utility. Ultimately, the future of AI will be built on a foundation of proactive, sensor-driven risk management that eliminates the invisible threats lurking in the pipes.
