As AI workloads escalate and data halls densify, what used to be โgood enoughโ for liquid cooling, no longer holds. As pressure builds (literally), infrastructure weak points start to show. And in too many facilities, that weak point is a leak.
Why Leak Risk Matters Now
In AI-scale deployments water leaks are a threat to uptime, safety, and reputation. The industry is racing to adopt high-density, liquid-cooled infrastructure, but with that comes a simple truth: as heat loads climb, flow rates rise. With higher flow comes greater pressure demands, raising the stakes for reliability and leak prevention.
Thatโs why forward-thinking infrastructure design needs to start with one question: what happens when something breaks?
How Our Leak Prevention System Works
Nautilus’ Leak Prevention System (LPS) is built around a patented design that enables sub-atmospheric operation in the data hall. Instead of pushing coolant through the entire system under positive pressure, we can pull it through under a controlled light vacuum. That simple inversion, pull instead of push, radically reduces leak risks. If a pipe or valve fails, air is pulled in rather than liquid pushed out. Nothing spills.
At the heart of this system is a Venturi powered by a motive pump. As water flows through the Venturi nozzle, it creates the suction effect that pulls coolant through the TCS (Technology Cooling System) loop in the data hall. This passive design avoids the complexity and fragility of mechanical vacuum pumps used by others, making it more resilient, more reliable, and less prone to failure.
Sub-Atmospheric When Possible, Leak-Preventative Always
Our Stockton facility has operated under sub-atmospheric conditions in LPS mode for years. Weโve seen zero leak incidents in the TCS, even in failure scenarios. Itโs as close to โleak-proofโ as you can get.
However, not all workloads allow for that. Modern high-density GPU racks, like NVIDIAโs latest platforms, introduce significant pressure losses that make full sub-atmospheric operation physically impossible. In these cases, we switch to hybrid variable pressure modes that maintain leak-preventative safeguards.
Our hybrid mode delivers precise variable pressure control, including positive pressure when required, and adapting seamlessly to the high flow demands of modern GPU environments.
With our Venturi-based design, the same mechanism that creates suction for leak prevention also acts as a built-in degasser. The Venturi coupled with our controls software forms the core of the Nautilus Air Detection and Removal System (ADRS). That system monitors and removes entrained air from the data hall loop, helping prevent bubbles, maintaining thermal stability, and preserving flow integrity even under extreme operating conditions. This integrated functionality eliminates the need for separate vacuum pumps or standalone degassing components, reducing system complexity and minimizing potential points of failure.
In short:
- Fewer moving parts = fewer failures
- Sub-atmospheric suction = true leak prevention
- Optimized differential pressure = broad support for high-density GPU systems
- Built-in degassing = resilience under pressure
Designed by Operators, Not Just Engineers
The LPS wasnโt designed in a lab. It evolved from real-world experience operating an AI facility cooled entirely by liquid-based infrastructure. Every design choice, from pipe routing to pressure tuning to filtration, was shaped by actual operating conditions, not theoretical models or slideware. When it comes to leak prevention, our operational experience matters. Because design decisions made in theory often break under pressure. Ours donโt.
Looking Ahead
As rack densities rise, pressure management will be even more critical. Weโre already working on next-gen configurations that extend LPS capabilities across facility-scale cooling architectures.
The industry is nearing a thermal breaking point. Intelligent leak prevention will go from โnice-to-haveโ to mission-critical. And the operators that invest in resilient cooling infrastructure today will be the ones still standing tomorrow.
