Boost your air-cooled server performance with an Industrial Two-Phase Passive Cooling Solution

Author: Olivier de Laet, Founder, Calyos; Frédéric Petit, R&I Director, Valeo Collection: 2026-OCP-EMEA-Summit-Executive-Sessions

Summary

Calyos and Valeo jointly present a passive two-phase loop heat pipe technology for direct-to-chip cooling in air-cooled servers. The technology uses a capillary structure — originally developed for the European Space Agency for satellite thermal management — to achieve pumpless fluid circulation. Vapor generated at the cold plate travels at up to 10 meters per second per kilowatt, condenses remotely, and is wicked back by the porous capillary structure without any pump. The physics advantage is the latent heat of vaporization: far greater thermal transport per unit fluid than single-phase water (which is already 3,500x better than air by specific heat), achieved passively and with a very large effective contact surface area.

Valeo contributes automotive-grade manufacturing scale and thermal management expertise. As a global Tier 1 automotive supplier with €20.9B in total sales, 100,200 employees, 59 R&D centers, and 149 production sites, Valeo brings expertise in two-phase heat pumps, radiators, cold plates, and battery cooling to the data center domain. CDUs and manifolds are already in production at Valeo's Zaragoza (Spain) plant for Zetacore. The partnership's first data center product targets the sub-50 kW rack market — which comprises 94% of racks sold in 2025 and is forecast to remain at 65% in 2030 — with a passive refrigerant-to-air system for HGX B300-class servers targeting commercial availability in 2027.

The Valeo-Calyos "Passive R2A" (refrigerant to air) concept applied to the Nvidia HGX B300 server reduces the server cooling height from 6U+2U (conventional heat pipe + air) to 4U+2U, cuts fan power by 80%, and enables 1,200 CFM airflow at a 68°C condenser case temperature versus 85°C for incumbent systems. The PoC specs include 8 cold plates (72×76 mm) covering 8 GPUs, 1,400 W per chip (11.2 kW server TDP), average heat flux of 46 W/cm² (max >100 W/cm²), fluid R-1233zd, and a 380×200×80 mm condenser. A 2U variant doubles chip TDP with the same approach. Future roadmap includes passive refrigerant-to-water (prototypes exist) and an active pumped variant to address higher power densities up to and beyond 150 kW per rack.