From Design to Operation: Commissioning Liquid Cooling Systems for Performance

Operators are working toward “time to tokens,” and commissioning now often happens under heavy time pressure. Flushing, purging and system preparation are being shortened or put off. Commissioning practices for liquid cooling are also still immature. Frameworks such as ASHRAE provide structure, but approaches still differ a lot from project to project.
Cutting validation short may get systems deployed sooner, but it moves unresolved issues into live environments, where they are harder and more disruptive to fix. In high-density AI environments, problems such as contamination, entrained air, incorrect configuration or incomplete integration rarely stay in one place. They lead to instability, weaker cooling performance and a slower path to stable, revenue-generating operation.
This is the second paper in the Liquid Cooling at AI Scale: Lessons from the Field series, developed by LiquidStack and Leading Edge Design Group. It walks through the five ASHRAE commissioning levels, from factory acceptance testing to integrated system testing under load. It also offers a clearer framework for more reliable operation and more predictable deployments.

What you’ll learn:

• Why skipping factory acceptance and witness testing doesn’t remove risk. It moves the risk to a stage where problems are harder to isolate and cost more to fix.
• How flushing, purging and prefilling during pre-start-up shape long-term system performance, and where this stage breaks down in practice.
• What Site Acceptance Testing should check at the TCS loop level, including flow balancing, coordinated CDU control and redundancy.
• How integrated system testing under load confirms thermal, hydraulic and control performance, including BMS integration, before production workloads go live.