Data centers
Hyperscale and colocation facilities running high-density AI racks.
Hayagreeva HEx
Engineered for next-generation AI accelerators — continuous heat removal at the semiconductor package, with a copper-matrix path from die to facility rejection.
HEx is the Hayagreeva Exchange — a direct-to-chip cold plate that mounts on the accelerator package. Fluid removes waste heat at the die and transfers it through the rack manifold and CDU to the facility heat-rejection system.
Package-specific fitting, mounting, connectors, and fluid are required for NVIDIA, AMD, and Intel accelerator architectures.
Continuous heat removal above 5,000 W at 0.005 °C/W — capacity stated at the package, not averaged across the rack.
Wetted surface density above 4,000 m²/m³ expands the heat-transfer interface at the chip boundary.
Chip → HEx → CDU → rejection. One path moves heat off the package and out of the building.
Rated thermal capacity and resistance are stated at the semiconductor package under defined test conditions. System-level performance depends on coolant properties, flow rate, inlet temperature, pressure drop, and package geometry.
| Thermal capacity | More than 5,000 W continuous heat removal (rated at package) |
|---|---|
| Thermal resistance | 0.005 °C/W rated |
| Architecture | Copper matrix with 4,000+ m²/m³ wetted surface area |
| Compatibility | NVIDIA, AMD, and Intel accelerator architectures |
| Intellectual property | 14 patent filings supporting the thermal design |
| Launch | May 28, 2026 |
Exact results depend on coolant, flow rate, inlet temperature, and package geometry.
Thermal performance is validated against stated test conditions. Rated metrics are separated from design targets. Contact engineering for NDA test documentation.
| Test methodology | Package-level thermal testing with documented coolant, flow, and inlet conditions |
|---|---|
| Claims policy | No unsupported superlatives; patent filings referenced as filings unless granted |
Hyperscale and colocation facilities running high-density AI racks.
Server and accelerator platforms requiring package-specific thermal integration.
Custom HPC and sovereign AI programs with facility-level cooling constraints.
Sustained accelerator performance without thermal throttling under load.
Open for qualified data-center operators, server OEMs, and integration partners.
Contact engineering