MHS Builds GPU Cold Plate That Cuts Coolant Demand

By Grace Kim

MHS has developed a cold plate for GPU cooling that runs on less coolant, the company announced in a report carried by Seoul Economic Daily. That single line matters for anyone building liquid-cooled AI servers, because coolant volume drives pumping power, plumbing cost and rack density.

A cold plate, for readers new to the term, is a metal block mounted directly on a hot chip — here a GPU — through which liquid circulates to absorb heat. It replaces or supplements air cooling in high-wattage accelerator setups. Less coolant per plate means smaller pumps, thinner lines and more compute per rack.

What does this change for data center operators?

For operators, the immediate question is coolant logistics. Direct-to-chip liquid cooling (a loop that touches the chip itself rather than the air around it) is already standard practice for flagship AI accelerators, and every reduction in fluid volume compounds across thousands of plates in a hyperscale hall. Fewer liters per plate cuts pump load and heat-rejection requirements at facility scale.

MHS positions the development within that shift. The Seoul Economic Daily report frames the product as a coolant-efficiency play rather than a raw thermal-performance claim, which distinguishes it from cold plate makers competing purely on maximum heat dissipation.

Why coolant volume is the metric to watch

The industry's cooling constraint has moved. It's no longer just "how many watts can the plate pull?" but "how much fluid and pumping energy does each watt cost?" Vendors that shrink the second number win design slots in next-generation racks, where space for manifolds and hoses is tight and every watt saved on cooling is a watt sold to the customer as compute.

MHS's development lands squarely in that argument. The company has not yet published full specifications — coolant flow rate, thermal resistance or compatible GPU packages — in the material reported so far.

What's next

Specification sheets and customer qualification results will determine adoption. Watch for:

• Flow-rate and pressure-drop figures versus incumbent plates
• Compatibility with current-generation accelerator packages
• Volume production timing and target customers

Until those numbers land, the headline fact stands: MHS says it can cool a GPU with less liquid. In a market where cooling is now a first-order cost line, that's a claim buyers will test fast.