For decades, the data center world has been divided neatly into two.
IT teams ran the servers; facilities ran the plant. If a rack overheated, facilities adjusted the CRACs and IT kept applications active.
That separation worked fine when racks drew 4-10kW.
But AI changed everything.
Today, a single cabinet can pull 120kW or more, and power and cooling have become inseparable. Data centers can no longer afford to operate like split departments. Every kilowatt, every water molecule, every line of code now depends on a unified system, and unified people, to keep performance stable and operations safe.
As Nautilus VP of Operations Gabe Andrews put it:
“We’re taking what used to be an entire data center and smashing it into a two-by-four-foot cabinet. That means the IT guy now has to know what a flange is.”
AI Is Collapsing the Divide
Air-based systems created a comfortable wall between IT and mechanical infrastructure. You could swap a server without touching a valve. If something failed, it was someone else’s domain.
Liquid cooling removes that wall entirely.
When water flows directly across chips, mechanical precision becomes inseparable from digital performance. Disconnecting a hose line, mis-reading a valve position, or skipping a pressure check can instantly trigger a cascade of temperature spikes and alarms.
In short: AI workloads aren’t just more powerful, they’re more intertwined.
And that interdependence is forcing a cultural shift inside the data center itself.
Living the Integration: Lessons from over 400,000 Unit Hours
At Nautilus, we’ve seen this evolution firsthand. Our Stockton facility, one of the only fully liquid-cooled AI data centers in operation, has logged over 300,000 unit hours just in that facility alone, 100K unit hours being live direct-to-chip performance. Our CDUs have 450,000 unit hours globally. Those hours have shaped not just our operational playbook but our product design itself.
Every EcoCore CDU we build reflects those lessons:
- Smarter control and monitoring: Our systems track every meaningful pressure change, temperature rise, and flow fluctuation, so the team sees problems before thresholds are crossed.
- Human-proof safeguards: Automated pressure equalization and variable-pressure Leak Prevention System (LPS) technology protect the loop even when human error occurs.
- Unified data visibility: Modbus, OPC UA, and SNMP integration connect mechanical and IT metrics into one intelligent control layer.
Operational experience isn’t just a credential; it’s a design principle.
We’ve lived through the challenges that most vendors have only modeled in CAD.
From Parallel Departments to a Single Organism
In many ways, the modern data center is evolving from a building into an organism, an environment where cooling, power, and compute constantly adapt to one another.
Historically, IT teams measured application uptime; facilities teams measured chiller efficiency. Now, those metrics are one and the same.
Every decision in the data hall, whether it’s a new GPU node or a valve adjustment, ripples through the same ecosystem.
That’s why our approach to infrastructure design always starts with integration. EcoCore systems don’t just deliver water, they deliver data. Every module communicates in real time, balancing electrical and mechanical loads across the entire facility.
The result? Fewer surprises, faster response, and true resilience under the extreme power densities AI brings.
The Rise of the Hybrid Engineer
This convergence is also changing the people who run data centers.
Tomorrow’s operators won’t just be mechanical engineers or IT admins, they’ll be hybrid engineers fluent in both domains.
They’ll know how to read a pump curve and a GPU utilization graph.
They’ll understand why valve sequencing matters as much as workload scheduling.
They’ll be the connective tissue between compute and cooling, the ones who keep both worlds in sync.
At Nautilus, that’s exactly how our operations team works. The collaboration between our product and field teams isn’t just culture; it’s necessity. Every innovation starts as an operational insight before it becomes a design spec.
The Path Forward: Unified Intelligence
The future of AI infrastructure won’t be defined by the biggest data halls or the fastest GPUs, it will be defined by how well cooling, power, and compute communicate.
Operators that embrace this integration will move faster, scale further, and run more efficiently. Those that don’t will find their facilities throttled by the same silos that once made them comfortable.
The age of separation is over.
AI demands convergence.
And at Nautilus, we’ve already lived it – 400,000 unit hours at a time.
Explore how Nautilus transforms real operational experience into product intelligence. Learn more about the EcoCore CDU product line and our variable-pressure Leak Prevention System.
