Technology
Two-phase jet impingement, directly at the die.
High-velocity dielectric jets impinge on the chip surface and boil on contact. Latent heat of vaporization does the thermal work: a fundamentally different mechanism from sensible-heat air or single-phase liquid loops.
Principle
Why two-phase, why jets
Boiling absorbs orders of magnitude more heat per unit mass than warming a liquid alone. Directing jets at the hottest region of the die keeps the boiling front stable and the thermal resistance low, even as heat flux climbs past what conventional cold plates can handle.
Figure — placeholder
Cold plate cutaway — schematic to be provided.
Comparison
Air vs single-phase vs two-phase jet-to-chip
| Approach | Heat transfer mechanism | Capacity limits | Coolant needed |
|---|---|---|---|
| Air | Fans blow air over a heatsink | Runs out of capacity around 500 W per chip. Already designed out of AI racks. | High fan power |
| Single-phase liquid | Water flows past the chip, absorbing heat and carrying it away | Works to about 2 kW per chip, after which the only lever is pumping unsustainable amounts of water. | About 1.5 litres a minute for every kW |
| Two-phase water (SemiValley) | Water boils on the chip surface. Boiling absorbs far more heat than warming. | Handles more heat per square centimetre than any chip expected in the next decade needs. | About 0.34 litres a minute per kW, roughly a quarter as much |
Detailed performance data available on request.
Performance That Speaks for Itself
<1.05
PUE
vs 1.3–1.6+ air
16×
thermal
vs single-phase
~$228M
annual savings
3 GW data center
Versus a modern single-phase liquid-cooled facility at PUE 1.15
10 kW+
TDP capacity
per chip
Who it's for
Built for the operators of AI infrastructure
Hyperscalers
Support next-generation accelerators at rack density without redesigning the mechanical floor around cooling.
Server & rack OEMs
Adopt a cold-plate that integrates into existing sled and manifold form factors.
Colocation operators
Lower PUE, recover stranded capacity, and offer high-density AI SKUs on existing facility loops.