Plan your deployment

Liquid Cooling Commissioning: What to Document from Day One

Liquid cooling has moved from pilot projects to the default plan for high-density AI racks. NVIDIA’s Vera Rubin platform, expected in data centers from fall 2026, uses warm-water, single-phase direct liquid cooling with supply temperatures of up to 45°C (113°F), at densities beyond the 120 to 130 kW per rack seen with current Blackwell systems.

That shift changes what commissioning has to prove. With air cooling, the main question at handover was whether the room holds temperature under load. With liquid cooling, there is a second system to hand over: the fluid itself, and every pipe, manifold, hose and sensor that carries it.

What gets documented in the first days of a liquid-cooled deployment decides how easy that system is to operate, troubleshoot and keep reliable for years.

Why temperature alone is not enough

Temperature is still one of the most important signals in a liquid-cooled facility. But it doesn’t tell the whole story.

In a September 2026 analysis for DCD, Rupesh Mainali of Reliability Engine describes how “a coolant can remain within thermal expectations while inhibitor reserve declines.” The loop looks healthy on a dashboard while it slowly becomes, in his words, “less tolerant of disturbance.”

By the time a temperature alarm fires, the underlying problem may have been building for months. Catching it earlier depends on having a baseline, and that baseline is set at commissioning.

The fluid needs a history

Mainali recommends keeping a “fluid biography” from commissioning onward: coolant type, chemistry measurements, filter changes, loop openings and service events, all linked to the equipment each loop serves.

He also points to an ownership problem. Coolant health “can fall between teams because it looks like chemistry to one group, facilities maintenance to another, and uptime risk to a third.”

Both problems are easier to solve on day one than in year three.

What to document, from first fill to handover

In our experience on the floor, these are the records that make the biggest difference once a liquid-cooled system goes live.

1. Design basis

  • Coolant type and specification
  • Design flow rates, supply and return temperatures, and pressure limits per loop
  • CDU setpoints and control logic
  • Which racks and equipment each loop serves

2. Installation and as-built records

  • As-built routing of manifolds, hoses and quick disconnects
  • Labels that match the drawings, at the rack and at the CDU
  • Clearances verified for maintenance access, especially on rear door heat exchangers

3. Flush, fill and pressure testing

  • Flushing procedure followed and results recorded
  • Fill date, volume and fluid batch information
  • Pressure and leak test results for each loop, with pass criteria

4. Baseline fluid chemistry

  • Chemistry readings taken at fill and again after the system stabilizes
  • Filters installed, with type and installation date
  • The baseline future samples will be compared against

5. Sensors and alarms

  • Flow, temperature, pressure and leak detection sensors listed by location
  • Calibration records
  • Alarm thresholds and who receives each alarm

6. Ownership and service log

  • A named owner for coolant health, agreed between IT, facilities and operations
  • A log for every loop opening, top-up, filter change and service event, starting at commissioning

None of this is complicated on its own. The challenge is doing it consistently across many racks while the schedule is compressing, which is why it works best when documentation is built as the installation happens rather than reconstructed at the end.

Standards are raising the bar

Documentation is also becoming a compliance question.

As Schneider Electric outlined in September 2026, California’s 2028 Title 24 cycle proposes new requirements for facilities with more than 2 MW of liquid-cooled IT, including economizer thresholds, minimum temperature differentials and variable-flow requirements. Adoption is expected in 2027, with the code taking effect on January 1, 2029.

At the federal level, H.R. 9372 would direct NIST and the Department of Energy to develop standardized ways to measure data center energy and water use.

A facility designed today may start operating under these rules. A clear record of how each cooling system was designed, installed and verified makes demonstrating performance much simpler.

The handover is the beginning, not the end

A liquid-cooled AI deployment isn’t finished when the racks power on. It’s finished when the operations team gets a system they can understand: what was installed, how it was tested, what normal looks like and who owns each part of it.

Getting that right at commissioning is one of the simplest ways to protect uptime later.

What does your team track from day one on liquid-cooled projects, and what do you wish had been recorded earlier? We’d like to hear what’s working on your floor. Reach out through weinstall.ai or connect with us on LinkedIn.

Sources

Scroll to Top