Two-Phase Liquid Cooling:            Your Questions Answered 

Liquid coolant selection in two-phase direct-to-chip cooling involves a careful balance of thermodynamic properties, materials compatibility, safety profile, and long-term regulatory outlook. Our fluid selection prioritises fluids with high latent heat of vaporisation, appropriate boiling points at operating pressures, chemical stability across component materials, and acceptable environmental and safety characteristics. We are not locked to a single fluid and actively evaluate the liquid coolant landscape as regulations and alternatives evolve.

This is an important and fast-evolving area. Some dielectric fluids used in electronics cooling — particularly certain fluorinated compounds — fall within PFAS regulatory scope depending on jurisdiction. We are actively tracking regulatory developments across the EU, US EPA, and other regions, and our fluid strategy accounts for near- and medium-term exposure. We advise customers on their specific deployment context and work to ensure our platform remains compliant and viable under foreseeable regulatory scenarios. Where fluid substitution becomes necessary, our system architecture is designed to accommodate transitions with minimal infrastructure disruption.

The fluid ZutaCore uses is non-flammable and has a low acute toxicity profile, consistent with it's established use in electronics and precision cooling applications. Handling protocols, materials safety data sheets (MSDS), and field service guidelines are provided as part of our deployment package. As with any engineered cooling system, ZutaCore provides training and documentation to ensure safe installation, operation, and servicing.

Natural coolants such as CO₂, propane, or ammonia present a range of challenges when applied to direct-to-chip electronics cooling — including flammability, pressure handling requirements, and materials compatibility at the chip interface. While these fluids perform well in certain macro-scale refrigeration cycles, the specific thermal and safety requirements of server-level two-phase cooling make them impractical in most deployment scenarios today. We continue to evaluate the landscape as both fluid chemistry and system architectures evolve.

 In ZutaCore's direct-to-chip architecture, the dielectric fluid contacts only the cold plate assembly — it does not flood the server chassis or make contact with PCBs, connectors, or other electrical components. This is a fundamental difference from full-immersion approaches. Signal integrity is not impacted. The thermal interface is contained and controlled, and server electronics operate in a standard, dry environment. 

ZutaCore HyperCool® operates as a sealed, pressure-balanced system. The fluid inventory is contained within the cold plate and CDU loop, and leak rates in properly installed systems are extremely low. In the rare event of a leak, the system is designed for controlled isolation and fluid recovery. Field repair procedures are documented and supported — our deployment teams and service partners are trained to respond and restore service. We do not expect leak management to be a routine operational burden.

Seal and fitting materials are selected for chemical compatibility with the operating fluid, thermal cycling resilience, and long service life. ZutaCore conducts lifecycle validation on all fluid-contact components as part of our product qualification process. Expected seal lifetimes are consistent with broader data center infrastructure refresh cycles. We publish component qualification data and can share detailed reliability documentation with customers under NDA for deeper evaluation.

Materials throughout the fluid loop — cold plates, tubing, fittings, CDU internals — are selected for chemical inertness with respect to the operating fluid. Two-phase dielectric fluids are typically non-corrosive and do not carry the ionic contamination risks associated with water-based systems. Erosion risk is managed through controlled flow velocities and appropriate materials selection. Our systems do not require the chemical inhibitor management that single-phase water loops demand.

ZutaCore CDUs use industrial-grade pump assemblies with MTBF ratings appropriate for continuous data center operation. Pump redundancy configurations are available for mission-critical deployments. In the event of a primary pump fault, the system design supports graceful degradation and rapid swap-out without full system shutdown, depending on configuration. We recommend discussing specific reliability SLAs with our engineering team during deployment scoping.

ZutaCore HyperCool is designed to minimise scheduled maintenance interventions. There are no water treatment regimes, no glycol management, and no strainer cleaning cycles. Routine checks focus on fluid level, pressure, and CDU health monitoring — all of which are available through the system's control interface. ZutaCore provides warranty, field service, and operations support models, including ZC-led operations for customers who want a fully managed cooling layer.

ZutaCore's cold plates are engineered for high heat flux direct-to-chip cooling, with capability aligned to current and next-generation GPU and accelerator silicon roadmaps. The two-phase approach — using latent heat of vaporisation rather than sensible heat — gives ZutaCore a fundamental thermal capacity advantage over single-phase solutions at equivalent flow rates and coolant temperatures. Specific heat flux ratings are available in our technical datasheets. 

Hotspot management is a core design objective for ZutaCore cold plates. The OmniTherm™ micro-structure uses engineered fin geometry and capillary wick structures to distribute liquid evenly across the die surface, continuously replenishing fluid at high heat flux zones. This enables stable, spatially uniform thermal control even under non-uniform GPU workload distribution — a common occurrence during real AI inference and training runs.

Critical Heat Flux is the point at which a boiling surface transitions from nucleate boiling (stable and efficient) to film boiling (unstable and thermally degrading). If reached, CHF causes a rapid and dangerous rise in surface temperature. ZutaCore's cold plate design operates with thermal margins well below the CHF threshold under all rated operating conditions. The micro-structured surface geometry, flow distribution design, and fluid selection collectively ensure that the system remains in the safe nucleate boiling regime across the full power envelope of supported silicon.

Vapour quality — the ratio of vapour to liquid in the two-phase mixture — is a key control variable in the system. ZutaCore's CDU and flow regulation architecture maintains vapour quality within a controlled band that ensures efficient heat transfer and prevents dry-out conditions at the cold plate. The system is designed to respond to changing GPU power draw dynamically, with flow rates adjusted to maintain thermal stability across transient workload conditions.

ZutaCore® HyperCool® is specifically designed for continuous, agentic, and real-time AI workloads — not just burst compute. The two-phase thermal mechanism inherently dampens power transients: the phase change process absorbs sudden increases in heat dissipation more smoothly than sensible-heat single-phase systems. This translates to lower GPU junction temperature variance, reduced throttling events, and more predictable performance at rack scale. Full-power, sustained compute is the design condition — not the exception. 

"Waterless" in ZutaCore's context means that water is not used inside the server or in direct contact with compute components. The HyperCool® system uses a dielectric fluid loop at the chip and rack level. At the facility level, dry cooler or adiabatic rejection options eliminate the need for cooling tower water entirely — making the full stack waterless from silicon to sky. In configurations where a facility water loop already exists, ZutaCore can interface with it at the CDU heat exchanger — but this is a facility-level choice, not a requirement. 

ZutaCore offers CDUs at both rack and row level, giving operators flexibility to scale cooling from a single rack to megawatt-class deployments.

In-rack CDUs install directly into standard 19" racks across four configurations — 20kW air-based (no facility water required), 60kW, 70kW, and up to 120kW with dual units — scaling linearly as density grows.

At the row level, the End-of-Row CDU family (1.2MW and 2MW) aggregates heat from multiple racks, condenses vapour back to liquid, and interfaces with facility water outside the white space, with N+1 redundancy across pumps, power, and controls.

Both levels integrate without changes to standard data centre infrastructure.

ZutaCore's platform has been developed with major OEM partners including NVIDIA, ASRock Rack, and Compal — with whom we have factory-integrated NVIDIA B300 systems as part of our qualified ecosystem. Rack and connector standards, including blind-mate fluid connections, are addressed within these partnerships. We recommend engaging ZutaCore's engineering team early in the rack design phase to confirm interoperability with your specific infrastructure configuration and standards requirements.

 ZutaCore CDUs include integrated control and monitoring systems that manage pump operation, flow regulation, pressure, and fluid temperature across the cooling loop. Key system health parameters are accessible via standard data centre management interfaces. The control logic is designed to respond autonomously to changes in GPU workload, inlet conditions, and system state — reducing the need for manual intervention during normal operation. Detailed control system specifications are available as part of the deployment documentation package. 

ZutaCore® HyperCool® is in active production deployment. We are powering a first-of-its-kind, fully waterless AI factory — demonstrating two-phase cooling at true AI-scale, not lab scale. Across 76+ global developments, our platform has been validated in real operational environments with enterprise customers and hyperscale-adjacent deployments. Factory-integrated systems with NVIDIA B300 servers are production-ready and available through our channel ecosystem. We are not a research technology — we are a deployment-ready infrastructure platform.