MHI’s 10MW chiller moves from test shipment to US market introduction

Mitsubishi Heavy Industries said on 19 October that its JHT-Z270I, a 10MW-class centrifugal chiller for AI data centres, had obtained ETL listing and AHRI certification. The company described the milestone as the start of market introduction activities for the United States.[1]

The sequence matters. In March, MHI announced its intention to enter the North American market, with commercial availability targeted after certification. That earlier plan described a single-compressor design and a larger plant architecture intended to combine efficient refrigeration with dry-cooler configurations that use less water.[3]

By July, the company had shipped a test unit to the United States. Its accompanying announcement described a Modular Chiller Plant, or MCP, integrating the chiller with pumps, heat exchangers and controls. It was being developed for scalable liquid-cooled AI facilities, including architectures aligned with Nvidia DSX.[2]

A large refrigerator in a larger system

A centrifugal chiller moves heat through a refrigerant circuit. Refrigerant absorbs heat in an evaporator, is compressed and then releases heat in a condenser before expanding and circulating again. MHI’s technical explanation of this cycle predates today’s AI build-out: a December 2010 engineering paper described it in a gas-turbine cooling application. The basic distinction remains useful here: cooling capacity measures heat removal, while the electricity needed to drive the compressor is a separate quantity.[6]

That distinction prevents a misleading reading of the headline number. A 10MW cooling designation describes the thermal job. It does not by itself say how much electrical power a completed installation will draw. Comparing the two requires an efficiency figure and the conditions under which it was measured.[6]

The modular approach addresses another job: joining equipment into a working installation. MHI says the MCP is designed to reduce on-site integration and support expansion. Supplying a coordinated group of components could make repeat deployments easier to engineer, although the receiving site still has its own operating temperatures, layout and cooling requirements.[2]

For an operator planning several construction phases, repeatability has value beyond the performance of an individual machine. A familiar arrangement can make the next phase a more bounded engineering task: decide how much additional cooling is needed, where it connects and how it will run alongside what is already there. That is the practical appeal of treating the plant as a product.

Two marks answer different questions

Performance also changes with the amount of work the chiller is doing. The US Department of Energy’s purchasing guidance distinguishes equipment optimised for full load from equipment optimised for partial load. It pairs requirements for both conditions because a machine selected for one operating pattern may spend substantial time in the other.[7]

This makes the utilisation pattern part of the engineering question. A cooling plant built ahead of server occupancy may begin lightly loaded; a later phase may operate much closer to capacity. The relevant comparison is the equipment’s behaviour across that pattern, rather than a single best-looking operating point.[7]

Let the weather do some of the work

MHI’s October description combines hybrid free cooling with a closed water loop. Outside air provides as much cooling as conditions allow, with the chiller operating when needed. The sealed circulation system is intended to limit contamination and evaporative water losses. The announcement supplies no measured whole-facility energy or water savings.[1]

The Department of Energy explains the general opportunity: suitable outside conditions can reduce reliance on mechanical refrigeration, while heat exchangers can let a cooling-water circuit remove heat before the chiller has to do the remaining work. The saving depends on climate, temperature settings and how many hours those conditions are available.[8]

Buying cooling at this scale means buying a repeatable engineering system. Its value lies in how reliably the parts work together as the computing installation grows.[1]

Report an error or suggest a correction