05Days
to OCP
Meet us in San Jose.

Two-phase cooling promises cheaper cooling through warmer water.

Two-phase cooling’s hidden pump bill

Published
Read time
2 minutes
Author
Phasic Energy

Most AI data centers being planned today run into the same problem: the chips make more heat than the building was designed to remove. Cooling has become one of the biggest costs of running a data center.

So we were glad to see The Register's recent look at two-phase cooling. Two-phase cooling pipes a special fluid straight onto the chips, where it boils off and carries the heat away, the same way sweat cools skin. The pitch from Accelsius, one of the companies building it, is simple: it lets a data center run its building water warmer.

Warmer water matters to the business. The colder a facility keeps its water, the more it spends on chillers, the large refrigeration units that are among the most power-hungry equipment on site. Accelsius estimates that each degree Celsius warmer saves about 4% in energy a year. Warmer water is also easier to reuse, for heating nearby buildings, instead of throwing that heat away.

We agree with the direction. Our question is about how you get there.

The fine print is the pumps

The warmest figures in the article depend on pushing twice as much coolant through the system. Pumps don't scale gently: in a typical loop, twice the flow can take roughly eight times the pumping energy. Some of the savings on the chiller side can come back as a bigger bill on the pump side, and the published numbers don't tell an operator how much.

Two-phase cooling still makes sense. A cooling system is only as efficient as its least efficient part, and the pumps are part of the system.

Where we come in

Every cooling system, two-phase or not, hands its heat to the building through a heat exchanger. That one component decides how hard the pumps have to work. Most heat exchangers are still built the way they were decades ago, from stamped metal plates, which forces a trade-off between moving heat well and letting coolant flow freely.

We 3D print ours, which lets us design the internal channels to do both. Our 1 MW heat exchanger, built for the Open Compute Project's Deschutes cooling design, is one example. A lower pressure drop across the exchanger means the pumps supply less pressure for the same flow, and draw less power. The exchanger is one part of the pump's load, so the saving shows up in that part of the bill.

What operators should ask for

The next time a cooling vendor quotes a water temperature, ask what it costs to pump. The industry's efficiency claims should cover the whole system, and we're holding ourselves to the same standard.

← Back to News

Upgrade to truly next-gen cooling architecture