ZutaCore and Munters Unveil Waterless Chip-to-Rooftop Cooling Architecture for AI Data Centers

New two-phase cooling design eliminates conventional CDUs and facility water loops, with modeled cooling energy savings of approximately 74%.

Key Highlights

  • Z₂O uses dielectric fluid and two-phase cold plates to transfer processor heat to rooftop equipment, where Munters SyCool rejects it to outdoor air.
  • The reference design is built around 625-kW IT blocks, with 500 kW of liquid cooling and 125 kW of air cooling.
  • The modeled pPUE comparison assumes Dulles, Virginia weather, constant full IT load and an 80/20 liquid-to-air cooling split.
  • ZutaCore says the first installation uses an ASRock Rack server with NVIDIA HGX B300, while independent operating results have not yet been reported.

ZutaCore has introduced a new waterless cooling architecture designed to simplify AI data center thermal management by extending two-phase liquid cooling from the processor to the facility's rooftop heat rejection equipment.

Unveiled at Supercomputing 2026 (SC26) in Chicago, the ZutaCore Z₂O architecture, developed in partnership with Munters, eliminates several components traditionally associated with liquid-cooled data centers, including cooling distribution units (CDUs), liquid-to-liquid heat exchangers, facility water loops and water-based cooling towers.

The architecture uses a sealed dielectric fluid that absorbs heat directly from processors through two-phase cold plates. As the fluid boils, vapor carries heat to rooftop equipment, where it condenses and returns through the same loop. Munters SyCool technology provides heat rejection to the outdoor air, eliminating the need for cooling water.

The approach represents an effort to address two increasingly significant constraints on AI infrastructure development: the availability of electrical power and the growing concern over water consumption.

Reducing the Energy Cost of AI Cooling

According to ZutaCore, the reference architecture developed with Munters achieves a modeled annualized partial power usage effectiveness (pPUE) of 1.03, compared with 1.115 for a conventional high-efficiency chilled-water system.

Those figures translate into approximately 74% less energy consumed by cooling infrastructure, based on the modeled difference in facility overhead.

The calculations use hourly weather data from Dulles, Virginia, assuming a constant 100% IT load and an 80/20 split between liquid and air cooling. The figures represent modeled performance rather than independently verified operating results.

Z₂O is designed around repeatable 625-kW IT blocks, consisting of 500 kW of direct-to-chip liquid cooling and 125 kW of air cooling. The modular approach is intended to support new construction, existing facilities and prefabricated data center deployments.

The architecture also incorporates N+1 heat-rejection redundancy and redundant UPS-supported pumps. ZutaCore's HyperCool Cloud software provides system monitoring.

"AI operators have fought hard for every megawatt, and they should not be spending it on moving heat," said Dan Wood, ZutaCore's chief product officer.

Implications for Data Center Development

For developers facing increasingly restrictive power allocations, reducing the electrical requirements of cooling infrastructure could free additional capacity for revenue-producing computing equipment.

Eliminating cooling towers, chillers and CDUs could also reduce mechanical infrastructure costs, simplify construction and shorten commissioning schedules. The extent of those benefits will depend on site conditions, system integration and actual operating performance.

The elimination of cooling water consumption is particularly relevant in markets where water availability has become a significant consideration in data center permitting and community acceptance.

ZutaCore said the first Z₂O installation uses an ASRock Rack server incorporating NVIDIA's HGX B300 platform with integrated two-phase cooling.

The architecture also supports ZutaCore's HyperCool and recently introduced OmniTherm cold plates, allowing future processor upgrades without requiring changes to the underlying facility cooling loop.

The announcement follows ZutaCore's $100 million-plus Series C funding round earlier this year, which included investments from Mitsubishi Electric, Carrier Ventures and Samsung Electronics.

The larger significance of Z₂O is its attempt to extend liquid cooling optimization beyond the server rack. While most direct-to-chip technologies focus on efficiently extracting heat from processors, ZutaCore and Munters are targeting the supporting mechanical infrastructure that contributes to overall facility energy consumption.

Whether the modeled performance advantages can be consistently achieved across different climates and operating loads remains to be demonstrated. Nevertheless, the architecture highlights the growing importance of treating cooling as an integrated facility design decision rather than an individual component of the IT infrastructure.

About the Author

DCF Staff

Data Center Frontier charts the future of data centers and cloud computing. We write about what’s next for the Internet, and the innovations that will take us there.
Sign up for our eNewsletters
Get the latest news and updates