immersion cooling vs direct to chip
Immersion vs direct-to-chip cooling
A head-to-head on single-phase immersion and direct-to-chip cooling: TCO, service model, density limits and dielectric fluid handling.
11 min read Updated 2026-05-30 Field-verified
Both technologies work. They serve different operating models. Pick DLC if you want to keep hot-swap workflows and OEM warranties; pick immersion if you're building a bespoke AI factory and can rewrite the service model.
One-line summary
Direct-to-chip is a plumbing upgrade to existing rack ops. Immersion is a new operating system for a data center.
Head-to-head
| Dimension | Direct-to-chip | Single-phase immersion |
|---|---|---|
| Density ceiling | ~150 kW/rack | ~250 kW/tank |
| Heat capture | 70–85% liquid | 100% liquid |
| Hot-swap | Standard | Drip trays + drain time |
| Warranty risk | Low (OEM DLC) | Case-by-case per OEM |
| Fluid cost | Low (water/PG25) | High (dielectric) |
| Weight/floor load | Rack + 100 kg | Tank + 1.5–2 t |
| Retrofit ease | High | Low |
When direct-to-chip wins
- You're upgrading an existing air-cooled hall.
- Your tenants demand OEM warranty coverage on GPUs.
- Ops teams need to keep familiar hot-swap procedures.
- You want fastest path to 60–130 kW/rack.
When immersion wins
- Greenfield AI factory optimised for a single silicon SKU.
- Extreme density (200 kW+/tank) with limited floor area.
- You are willing to run a dielectric chemistry program.
- Silent operation is a hard requirement.
FAQ
Can I mix both in one hall?
Yes, but plan the loops separately. Cross-contamination between glycol and dielectric fluids is expensive to remediate.
Need immediate hydronic support?
Critical leak response, pump diagnostics, and pressure stabilization. Engineers en route in under 4 hours.