MHS Claims Cooling Plate Cuts AI GPU Coolant Use by 60%

By Priya Raman

MHS says its new cooling plate for AI GPUs reduces coolant consumption by 60 percent. The figure comes from the company itself. No independent test data accompanied the announcement, and the publication date of the claim was not stated in the release circulated by Asia Economic Daily.

Cold plates are metal blocks with internal channels that carry liquid directly to a chip's heat spreader. They matter because AI GPUs, such as those used in large training clusters, draw several hundred watts each and push air cooling past its practical limit. Coolant flow rates, pump power and piping volume are now real cost items for data centre operators.

A 60 percent reduction in coolant use, if verified, would cut both pump energy and fluid volume in liquid-cooled racks. That is the promise. The announcement does not say which GPUs the plate fits, at what thermal load the figure was measured, or whether the reduction applies to flow rate, total fluid inventory, or both. Each of those readings implies a different saving.

What is not in the announcement?

The company did not publish:

• The test conditions or thermal load used to derive the 60 percent figure
• A release date or volume production schedule
• Pricing, partner names, or target customers
• Comparative data against incumbent cold plate suppliers

Vendor-measured efficiency claims in thermal hardware routinely shrink under third-party testing, because real racks impose flow restrictions, coolant mixtures and maintenance constraints that bench setups omit. Until MHS publishes measurement conditions, the 60 percent figure functions as a marketing claim rather than a specification.

Why coolant volume matters now

Liquid cooling has moved from niche to standard for high-density AI compute. Racks that once drew 20 kilowatts now exceed 100 kilowatts when packed with accelerators, and each additional litre of circulated coolant per minute costs pump power, plumbing and leak risk. A supplier that genuinely cuts flow requirements by more than half would relieve a genuine constraint, which is precisely why such claims deserve scrutiny before procurement decisions rest on them.

What should buyers ask?

Any data centre operator evaluating the plate should request the thermal resistance curve (the measure of how effectively the plate transfers heat at a given flow rate), the pressure drop across the plate, and the fluid type assumed in testing. Without those three numbers, a coolant-reduction percentage cannot be compared against competing products on equal terms.

MHS has not yet responded publicly with this data. We will update this article when measurement details, certification or customer trials surface.