Direct-to-chip liquid cooling (single-phase) delivers coolant directly to the highest heat-generating components – such as CPUs and GPUs – using cold plates.
Unlike immersion cooling, the liquid does not come into direct contact with electronic components, allowing servers to retain familiar form factors and serviceability.
How It Works
- Coolant flows through cold plates attached to high-power components (e.g., CPUs, GPUs, memory, and power delivery elements)
- Heat is transferred to a liquid loop
- Residual heat is managed with reduced airflow
Why It’s Preferred
- Targeted cooling of high-heat components
- Easier integration with existing server designs
- Supports retrofit and hybrid environments
- Maintains standard rack-based architecture
Key Trade-Off
Some airflow is still required to cool non-chip components. In practice, this often results in hybrid architectures, where liquid cooling handles high-power components while existing air systems – such as CRAHs or fan walls – manage residual heat within the room.
Key Takeaway
Direct-to-chip cooling offers the best balance of performance, efficiency, and compatibility for most AI deployments.
