Daikin Applied Unveils 60 MW D60-W Cooling Design for NVIDIA DSX Racks

Daikin Applied introduced a 60 MW cooling reference architecture for NVIDIA-based AI factories on Nov. 10, 2026, with liquid cooling carrying 92% of the IT load and air the other 8%.

The design, called D60-W, is a blueprint for the NVIDIA DSX AI Factory Platform with Vera Rubin NVL72 racks, according to Daikin Applied's release and its reference-architecture page. It supports a 60 MW IT load across six deployable units, built around the company's oil-free Magnitude WME magnetic-bearing centrifugal chillers, which come in capacities up to 10 MW. HPAC Engineering carried the same specifications.

The point for operators is water and power. D60-W rejects heat through dry coolers rather than evaporative cooling towers, so it needs no cooling-tower make-up water, and Daikin says the dry coolers carry the load for most of the year in suitable climates. The chillers cover the hottest days. That is Daikin's claim for a reference design, not a measured result at a site, and the company gives no water usage effectiveness (WUE) or power usage effectiveness (PUE) figure.

How the design splits the load

Each deployable unit has a reference capacity of 13.8 MW, with 10 MW allocated to the plant. Daikin's page breaks a unit into 9.2 MW of liquid cooling and 0.8 MW of air cooling, which is the 92% and 8% split, and lists 128 racks across four scalable units. Seven 2.3 MW coolant distribution units (CDUs) serve each unit, six on duty and one on standby. DDC Solutions S-Series rack containment units (RCUs) capture the remaining heat at the rack.

Two temperature loops run side by side. A high-temperature loop serves the CDUs, and a low-temperature loop serves the RCUs. On the reference page, the facility liquid loop runs at 37°C supply and 47°C return, and the rack liquid loop at 41°C supply and 51°C return. The RCU water loop runs at 22°C supply and 33.11°C return. The design weather basis is Ashburn, Virginia, with a 39.8°C summer dry bulb.

Three operating modes

In dry mode, the chiller compressors are off. It applies when the dry cooler leaving-fluid temperature plus the heat-exchanger approach is at or below 41°C. During full economization only, facility supply resets up to 41°C, which Daikin says provides 45°C rack supply through the CDU. Between 41°C and 47°C, the hybrid mode uses waterside economizers to precool the return water, and the chillers handle the rest. At 47°C or above, mechanical cooling takes over.

Daikin's iDCM supervisory layer coordinates the RCUs, CDUs, chillers and dry coolers as one plant, handles load sharing and manages the mode changes. Central plant redundancy is N+2: six plus two quad WME chillers and 21 plus 2 dry coolers on the high-temperature loop, and four plus two WME chillers and two plus two dry coolers on the low-temperature loop. The page adds that the heat-rejection plant supports 60 MW of IT load and that more plant capacity is needed to use the full thermal capacity of the six units.

Where it sits in the market

Reference designs are becoming the unit of sale for AI cooling. Vertiv bundled liquid cooling and power for GB300 and Rubin racks, as Voltgrid reported, and FläktGroup built a 2.5 MW CDU for a 25 MW US data hall. Daikin's 2.3 MW CDU is in the same size class, but the Daikin-branded CDU is not yet a product.

What is not confirmed yet

Daikin says its own CDU and data center dry cooler are under development, with availability targeted for the second half of 2027. The design is a reference, not an installed plant, and no customer, site or project is named. No efficiency, WUE or PUE numbers are published, and the share of the year in dry mode depends on the climate. Whether NVIDIA has validated or endorsed the design is not stated; the release says only that it is based on the DSX platform and Vera Rubin NVL72.