
Power, not silicon, has become the ceiling on AI, and a surprising share of that power goes to moving heat. On November 16, 2026, at Supercomputing 2026 in Chicago, cooling specialist ZutaCore unveiled Z₂O, a waterless cooling architecture that it says collapses the entire thermal stack into a single loop, according to the company’s Business Wire release.
One loop from chip to rooftop
Z₂O runs a unified two-phase loop from ZutaCore’s direct-to-chip cold plates all the way to rooftop heat rejection. ZutaCore says that means no coolant distribution unit (CDU), no liquid-to-liquid heat exchanger, no facility water loop and no water-based cooling towers. Dry coolers reject the heat to outdoor air instead.
The launch partner is Munters, whose SyCool heat rejection technology completes the design. In the reference design the two companies modelled, the result is an annualized pPUE of 1.03 with zero water use.
Why two-phase makes it possible
The trick is physics. In two-phase cooling, a dielectric fluid boils on the chip and condenses on the roof at essentially the same temperature, so heat moves at a constant temperature without intermediate stages. ZutaCore says the fluid is sealed, non-toxic, non-flammable and non-conductive, and circulates largely by buoyancy and gravity, with pumps for assistance and redundancy.
Because one loop does the work of the whole chain between chip and outdoor air, the equipment that normally consumes energy and water, including the CDU, heat exchanger, cooling tower and chiller, drops out of the design. ZutaCore says that is where both the energy and water savings come from.
The numbers, and their fine print
An annualized pPUE of 1.03 means roughly 97 percent of the power secured for IT and cooling reaches the IT equipment, per the company. In the same model, ZutaCore says best-in-class chilled water reaches 1.115, so Z₂O uses about a quarter of that design’s cooling energy.
These are modelled results, not measured ones. The footnote specifies hour-by-hour Dulles, Virginia weather data, a 43 °C saturation temperature, an 80/20 liquid-to-air split and 100 percent IT load, following ASHRAE 2026 methods. Results in a hotter or more humid climate, or at partial load, could differ.
Zero facility water also means a water usage effectiveness (WUE) of zero for both the liquid-cooled and air-cooled load, which could open sites in water-constrained regions.
Built from repeatable blocks
The architecture is organized around a repeatable 625 kW IT block, made up of 500 kW of two-phase liquid cooling plus 125 kW of air cooling. ZutaCore says this suits data halls, prefabricated modular data centers, and both greenfield and brownfield sites, with fewer systems to design, permit, install and commission. Spare heat-rejection capacity (N+1), redundant UPS-fed pumps and fewer active systems are meant to keep operations running if one part fails, and the HyperCool Cloud software gives operators visibility across the loop.
Z₂O is built to support servers from multiple manufacturers. The first installation runs on ASRock Rack’s NVIDIA HGX B300 server, described as the first with fully integrated waterless liquid cooling. It is also compatible with the OmniTherm two-phase matrix flow boiling cold plates unveiled at the OCP Global Summit in October 2026, and with the existing HyperCool plates, so customers can upgrade the cold plate without re-plumbing the facility. ZutaCore says more partner implementations will follow in 2027.
What it means
AI operators are increasingly limited by grid connections and by local objections to water use. A design that returns more of each secured megawatt to compute, and that needs no water, speaks to both problems. Skipping the chiller plant, towers and CDUs could also lower capital and maintenance costs, a point ZutaCore makes about total cost of ownership.
The caution is that this is an announcement of an architecture and a first installation, backed by vendor modelling. Operators will want measured pPUE from live deployments, climate-specific analysis, and clarity on dielectric fluid supply and handling before committing at scale. Still, ZutaCore says its technology has been proven across more than 75 deployments, and Z₂O extends that approach to the whole facility.
Conclusion
As chips on industry roadmaps climb toward multi-kilowatt power levels, cooling is becoming a headline design choice rather than a back-room detail. Z₂O is a clear statement that the next gains may come from rethinking the building around the chip, not just the cold plate on top of it.