MDCX solutions are factory-built containers in two lines: Liquid Cooling and Immersion Cooling. Each block has a readable name and a code name — start with TCO before you lock a configuration.
Model CapEx, power, utilization, and IRR for Liquid Cooling and Immersion Cooling — then open Designer with the same solution code.
Cold-plate containers for high-density training racks. Factory-built, one warm-water or dual-loop interface.



Single-phase immersion containers for inference density and a lower cost per GPU on the pad.



Pick the line by platform, then the container by pad length and UPS boundary.
| Liquid Cooling | Immersion Cooling | |
|---|---|---|
| Cooling method | Cold plate on the GPU | Single-phase bath of the whole board |
| Platforms | SXM / NVLink (GB300 NVL72 · B300 HGX) | PCIe fleets |
| Density | Up to 220 kW/rack | Up to 50 kW/tank |
| PUE | ≈ 1.15–1.20 typical mid-climate only — not free-cooling or summer-peak; confirm in TCO / Designer | ≈ 1.05 · peak dry-bulb ≤ 24 °C / free-cooling — not hot-zone |
| IT capacity | 450 – 1,800 kW | 200 – 400 kW |
| Typical load | Training | Inference |


Standardized telemetry, alarms-to-tickets, operations workflows, and usage metering — from a single container to a full cluster. Open-core Apache-2.0 at github.com/MDCX-ORG/CIOS.