Thermal Intelligencefor the Age of AI Power
Scalable two-phase cooling.
Engineered for deployment.
Making Two-Phase Cooling Deployable for Next Generation AI.
PhaseQ starts where the demand is urgent: high-density AI data centers, where power, water, space, cooling, and deployment limits constrain compute growth. We build the thermal intelligence that makes two-phase cooling deployable as reliable infrastructure.
STAR is our first tangible architecture for this mission. It is designed for 250+ kW/rack, CDU-compatible integration, low water demand, and reliable operation under transient AI workloads.
Thermal Limits Should Never Be the Ceiling On Human Ambition.
PhaseQ is building the deployment layer for frontier thermal infrastructure — starting with high-density AI compute and expanding into advanced energy systems.
Cooling performanceis not the bottleneck.Deployability is.
AI compute is outgrowing the deployable limits of conventional cooling infrastructure. The limiting question is no longer only how much heat can be removed at the chip, but how much cooling performance can be integrated, operated, and serviced at rack and cluster scale.
Two-phase cooling offers the thermal performance needed for this transition, but performance alone is not enough. Without rack-level architecture that stabilizes two-phase behavior and fits existing infrastructure interfaces, high-density cooling can become difficult to deploy, validate, and maintain.
A Two-Phase Architecture for Deployable 250 kW+ AI Racks
STAR is PhaseQ's first tangible architecture for deployable two-phase AI cooling. It sits between the server heat load and CDU-side infrastructure, passively conditioning phase behavior to smooth vapor quality fluctuations, pressure response, and workload-driven transients.
STAR is designed to turn high-density two-phase cooling into practical AI compute infrastructure — targeting 250+ kW/rack deployment, CDU-compatible integration, no facility-side redesign, simplified maintenance, and reliable operation under real AI workloads.
Deployable Capacity
Converts phase-change cooling performance into rack-level capacity by conditioning the two-phase boundary between accelerator heat loads and CDU infrastructure.
System-Scale Stability
Damps vapor-quality, pressure, and thermal excursions driven by transient AI workloads before they propagate through the cooling loop.
Passive Deployment Interface
Designed as a CDU-compatible thermodynamic layer with no active valves or control loops inside the STAR boundary, simplifying integration and maintenance.
Cooling Performance Is Not Deployable Capacity
Two-phase cooling can remove intensive heat at the chip, but AI data centers need that performance to translate into rack-level infrastructure capacity. Vapor-quality excursions, flow maldistribution, pressure dynamics, maintenance complexity, and facility-integration constraints can turn theoretical cooling capability into stranded deployable capacity.
A Thermodynamic Layer Between Chip and Infrastructure
STAR inserts a passive phase-conditioning layer at the rack boundary. By stabilizing the two-phase return path and presenting a more manageable interface to CDU infrastructure, STAR is designed to help high-power AI racks deploy faster, operate more reliably, and scale through a no-facility-redesign integration path.
Architecture-First.Interface-Ready.
High-power AI racks need more than chip-level cooling; they need rack-level architecture that turns chip-level cooling into deployable capacity.
PhaseQ develops STAR to directly address the infrastructure bottlenecks that limit cooling power, rack capacity, and deployment speed.
Shift the Thermal Regime
Cooling power becomes a bottleneck when heat loads outgrow facility-side cooling capacity. STAR uses phase-change heat transfer at the rack-to-chip boundary to carry higher heat loads while easing facility-side power and water demand.
Stabilize the Rack Boundary
Deployable capacity is lost when two-phase performance cannot remain stable at rack scale. STAR conditions vapor quality, pressure behavior, and workload transients so two-phase performance becomes usable capacity for 250+ kW/rack targets.
Standardize the Interface
Deployment speed slows when every site requires custom engineering. STAR acts as a passive CDU-compatible interface, reducing custom integration, validation, service, and maintenance complexity.
Building the thermal deployment layer
AI compute infrastructure needs.
PhaseQ is developing rack-to-chip thermodynamic architectures that expand deployable AI compute capacity — bridging two-phase heat transfer physics with infrastructure-compatible deployment.