Why the immediate future of liquid cooling is hybrid
The article examines how AI training and inference workloads are driving thermal management changes in data centers. Traditional air cooling is becoming insufficient as GPU-heavy AI racks operate between 40 kW and over 100 kW, with next-generation systems approaching 200 kW per rack. This density surge makes hybrid cooling—combining air and liquid systems—essential rather than experimental, forcing operators to rethink infrastructure architecture and deployment timelines.
Key points
- Air-only cooling is viable only up to 20-40 kW per rack; modern AI racks operate at 40-100+ kW.
- GPU-driven density has made hybrid liquid and air cooling mandatory for AI and HPC workloads.
- Industry focus has shifted from 'if/when' to rapidly deploying and scaling hybrid systems.
Summary and key points written by AI from the headline and a short excerpt, so they can be incomplete or wrong. The Register did the reporting; read the original for the full story.
Read the full story at The Register
