CoolIT says near‑total liquid heat capture enables fanless AI servers beyond 250 kW
Hybrid liquid‑air cooling works up to about 250 kW per rack, but a 70/30 split leaves roughly 75 kW of air load. CoolIT argues that moving that heat to liquid makes the air component drop below one percent, allowing servers to run without fans and avoiding the cost and footprint of parallel air systems.
Key points
- Hybrid liquid‑air cooling fails above 250 kW per rack, leaving ~75 kW air load.
- CoolIT’s modular coldplate blocks aim for 100 % heat capture, enabling fanless servers.
- Modeling projects full heat capture as standard for rack‑scale AI products through 2028 as power rises toward 1 MW.
The company builds modular cold‑plate blocks that can be combined into a single loop covering CPUs, GPUs, memory, networking, storage and power modules. Its toolkit includes conductive plates, vapor chambers, heat pipes and thermal transfer plates, each matched to the shape and thermal limits of the component. CoolIT says the loop remains easy to install and reliable, with connection and routing design that scales to higher densities.
CoolIT’s modeling projects full heat capture to become the default rack‑scale design for flagship AI products through 2028 as rack power climbs toward 1 MW. Operators interested in fanless, fully liquid‑cooled servers are invited to discuss custom loop designs with the vendor.
The Future Is Fanless: 100% Heat Capture for Liquid Cooled AI Servers
IEEE Spectrum AI · 22 September 2026
Loading the full article…
This text was published by IEEE Spectrum AI and written by Coolit, an Ecolab Company. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Hardware & Compute
All →- Opinion: M5 Ultra Mac Studio enables fast local AI agents, reviewer says · 2 src
- Google adds 400‑frame Magic Capture to Pixel 11 camera · 1 src
- Jeff Dean says automation could cut chip design teams to ten and cycles to three months · 1 src
- Mixture of experts reshapes inference compute and data movement pipelines · 1 src
- Nvidia launches DSX Ready program to qualify power and cooling products · 1 src
Comments
via GitHub Discussions