AI Data Centers Get 92% Liquid-Cooled Blueprint

Daikin's 60 MW D60-W design cools most load with liquid, a reference architecture announced but not yet shipped.

Updated on Nov. 14, 2026 in Data Centers

Isometric editorial illustration of two stacked teal cooling cabinets linked by pipe runs beside a row of louvered dry-cooler housings on a flat pad. — Daikin Applied on November 14 announced the D60-W, a 60 MW cooling reference design for AI data centers that handles most heat with liquid, though it has not yet shipped. AI Illustration. Upload story photo >

Daikin Applied announced on November 14, 2026, a 60 MW cooling reference architecture for AI data centers, called the D60-W, built on the NVIDIA DSX AI Factory Platform with NVIDIA Vera Rubin NVL72. The architecture is announced rather than shipped, and the extraction does not indicate availability dates.

Why it matters

AI facilities are running denser racks that generate heat in volumes conventional air cooling struggles to remove, so cooling design is becoming a limiting factor in building them. The D60-W reflects a shift toward liquid cooling as the primary method, with air cooling reserved for a small remainder of the load.

The specification is a manufacturer reference design, so the 60 MW, six-unit, 92/8 liquid-to-air split describes intended capacity rather than measured operation. The architecture pairs oil-free magnetic-bearing centrifugal chillers rated up to 10 MW with dry coolers in place of evaporative towers, and no independent benchmark or efficiency comparison is provided in the source data.

The players

Daikin Applied

Daikin Applied is a commercial HVAC manufacturer whose data-center portfolio includes the Magnitude WME oil-free magnetic-bearing centrifugal chillers used in the D60-W design.

NVIDIA

NVIDIA is a GPU and AI-computing company whose DSX AI Factory Platform and Vera Rubin NVL72 rack system the D60-W architecture is built around.

The details

The D60-W runs two separate cooling loops: a high-temperature liquid loop serving coolant distribution units that carry heat away from the chips, and a low-temperature loop serving air-cooling rack containment units for the remaining 8% of load. Heat is rejected through dry coolers rather than evaporative cooling towers, and the system shifts between dry, hybrid and full mechanical cooling depending on outdoor conditions. In dry mode the chiller compressors are switched off; in hybrid mode waterside economisers pre-cool return water so the chillers handle only the leftover load. A supervisory layer called iDCM coordinates the containment units, distribution units, chillers and dry coolers as a single plant.

November 14, 2026: Daikin Applied announces the D60-W cooling blueprint.

The Tech Race

The D60-W is positioned as a cooling layer for the NVIDIA DSX AI Factory Platform, which pairs the NVIDIA Vera Rubin NVL72 with a reference facility design. The extraction does not name competing cooling architectures, so the competitive comparison is limited to the platform pairing itself.

The architecture is a reference design, and the extraction gives no availability date, pricing, or customer deployment, so its effect on users is not yet determined. Operators planning AI facilities around NVIDIA Vera Rubin NVL72 racks would be the first to evaluate a design like this.

The takeaway

Watch for the first named deployment of the D60-W, since a reference design becomes meaningful only once a facility runs it at scale. The extraction gives no date for that, so the next verifiable milestone is a customer or operator announcement.