
AI servers are getting hotter faster than most server rooms can cope. Valeo, the French automotive supplier, used Data Centre World Paris on November 18 and 19, 2026 to show how it plans to help: an aluminum rear door cooling module, direct-to-chip cold plates, in-rack cooling distribution units and a pump-free loop heat pipe system. Here is what the company announced and why the details matter for anyone running dense compute.
An aluminum rear door instead of copper
According to Valeo’s press release, the new Rear Door Cooling Module (RDHx) swaps the traditional copper coil for aluminum. Valeo says that change delivers a 38% boost in cooling capacity and a 65% increase in power density compared with conventional copper equivalents. It also cuts electrical fan consumption by 25%.
The module is built for high-density racks and fan wall applications across several rack sizes. A proprietary surface coating is meant to protect the aluminum for the long term. Valeo says the design lowers both PUE (Power Usage Effectiveness) and CUE (Carbon Usage Effectiveness), and that it suits new builds and retrofits alike, including cloud racks and air-cooled AI inference racks.
Rear door heat exchangers sit on the back of a rack and pull heat out of the exhaust air before it reaches the room. That makes them one of the easier ways to add capacity to an existing hall, because the servers themselves do not change. Valeo’s pitch is that a lighter, lower-copper coil can do more of that job with less fan energy.
Cold plates and CDUs for the hottest chips
For direct-to-chip cooling, Valeo is displaying both single-phase and two-phase cold plates for CPUs and GPUs. The company describes a multi-material architecture with a no-seal casing saddle assembly process, which it says minimizes copper use and maximizes leak resistance. The internal heat-exchange design can dissipate heat fluxes of up to 200 W/cm² while optimizing pump mass flow, according to the release.
Alongside the plates, Valeo shows Cooling Distribution Units (CDUs) and manifolds for single-phase or two-phase use. It says they are built to automotive quality standards, with hot-swappable pumps, brazed junctions, failure prediction controls and predictive maintenance software. Those are the features operators tend to ask about first, since a pump failure in a liquid loop is an operations problem rather than a design curiosity.
A passive loop heat pipe for air-cooled servers
The third piece targets servers that will stay on air. Valeo notes that 47% of AI servers are air cooled today, and says even the best air-cooled servers cannot run today’s top GPUs at full power. Co-developed with Calyos, its demonstration is a passive two-phase direct-to-chip system at server level based on Loop Heat Pipe (LHP) technology.
Instead of mechanical pumps, the loop relies on capillary and gravity action. Valeo reports 27% higher cooling capacity than air, thanks to the two-phase fluid, and a 40% reduction in fan consumption. The company also positions it for edge compute, where a self-contained cooling approach with no pump to maintain could be attractive.
Why an automaker’s supplier is in the data center
Pascale Herman, Valeo’s Data Centers and BESS Vice President, said in the release that data centers are caught between rising AI compute density and strict energy targets. Valeo says it is bringing automotive-grade reliability and industrial scalability to the sector, drawing on decades of thermal engineering and high-volume manufacturing.
The logic is plausible. Cars already pack heat exchangers, pumps and coolant loops into tight spaces and must survive years of vibration and temperature swings. Skills from that world transfer naturally to compact, mass-produced cooling parts, which data center operators increasingly need in volume.
What it means
Three takeaways stand out. First, the figures are Valeo’s own comparisons and the release does not publish test conditions, so buyers will want independent validation before counting on them. Second, the portfolio covers a spectrum: rear doors for retrofits, cold plates for the newest accelerators, and passive loops for servers that cannot move to full liquid cooling. Operators rarely switch everything at once, so a mix is realistic. Third, the emphasis on fan energy shows where the pressure lies. Cutting fan power by 25% to 40% reduces facility overhead directly, which feeds the PUE and carbon metrics operators now report.
The release does not list pricing, availability dates or named customers, so this is best read as a product showcase rather than a deployment announcement.
Conclusion
Valeo’s Paris showing signals that thermal management for AI is becoming a crowded field, with automotive suppliers now competing alongside traditional data center vendors. If the claimed capacity and fan-energy gains hold up in customer testing, aluminum rear doors and passive heat pipes could give operators practical ways to squeeze more compute from existing halls.