Platform library
What an existing hall is actually being asked to carry.
Every figure below comes from the manufacturer’s own published material and is used unchanged by the engine that screens halls. Vendor-sourced values are marked as such in the provenance of every study they appear in.


| Platform | Rack kW | Peak | Liquid | Residual air | Max inlet | Flow | Floor | Feed |
|---|---|---|---|---|---|---|---|---|
| NVIDIA GB200 NVL7272 GPUs · shippingQCT QoolRack GB200 NVL72 technical brief, 2025 [V] | 132 | 132 | 87% | 17.2 kW | 45 °C | 125 l/min | 1,889 kg/m² | 200 A |
| NVIDIA GB300 NVL7272 GPUs · shippingLenovo Press LP2357, GB300 NVL72, 2026 [V] · NVIDIA NVL72 AI Factory Reference Architecture, 2026 [V] | 135 | 155 | 90% | 13.5 kW | 45 °C | 130 l/min | 1,944 kg/m² | 205 A |
| Generic DLC server rack, 50 kW32 GPUs · shippinggeneric archetype — no manufacturer, no source | 50 | 55 | 75% | 12.5 kW | 40 °C | 72 l/min | 1,250 kg/m² | 76 A |
Residual air
is not zero
Liquid cooling shrinks the air load; it does not remove it. The remainder still needs the plant the hall already has.
Floor loading
against ~1,220 kg/m²
A TIA-942 Rated-3 raised floor design value. Two of these platforms exceed it distributed, before any point load at the castors.
Max inlet
sets your approach
The gap between your facility water and this figure is the approach a CDU actually has to work at — rarely the 5 K on the datasheet.
Deliberately absent
Platforms whose rack power the manufacturer has not published are absent by design rather than pending. An invented number in a capacity study is how an engineering reputation ends.
- vr200_nvl144
- NVIDIA has not published rack power. Third-party estimates 190-230 kW are unverified.
- rubin_ultra_nvl576
- ~600 kW is a stated target from GTC reporting, not a specification.
- amd_helios_mi450
- AMD has not published rack-level power for MI450/MI455X.
Can your hall take one of these?
Pick the platform, give seven numbers about the hall, and the engine names the constraint that stops you first. Free.