ORBIS ELECTRIC
O | E
100 MW IT AI SCALE REFERENCE POINT
Meeting NVIDIA at the
Chiller-less Goal Line in 2026
How Cool-E™ Potentially Reallocates 3+ MW to IT / GB200-Class

Pumps, MCA, and 238 GPM/motor flow are locked identically across both generations ‚Äî only the required HX area and achievable approach change, because the two NVIDIA loop specs impose different capacity-rate balances on the same hardware. Every number below passes the Qmax = Cmin×(Thi−Tci) energy-balance check before being displayed.

DC-native 800V DC bus, no rectifier — 27A
AC-retrofit 480V AC + power supply front-end — 37A

Rack Reallocation GB200, 132kW/rack

GB200 NVL72 rack power: 132 kW/rack (fixed — QCT QoolRack spec: 115kW liquid-cooled + 17kW air-cooled, not adjustable)

Ground Zero Legacy CDU, real spec

The actual hardware Cool-E retrofits: 1.2 MW, 5K approach (35‚Üí45¬∞C facility / 40‚Üí50¬∞C IT), 234-238 GPM/motor (2+1 Grundfos CME25-2, 7.5HP induction), 63A cabinet breaker (printed spec, whole-cabinet incl. controls). Its 35¬∞C max inlet sits between ASHRAE TC 9.9's W32 and W40 facility-water classes (no official "W35" class exists) ‚Äî representative of the pre-AI-boom mainstream, not a fringe spec. Field-observed: motors run up to 120% of nameplate , drawing as much as 21kW each (42kW/2 motors) — this is why 238 GPM/motor is fixed hardware, not a Cool-E design choice: Cool-E retrofits the same chassis, same flow, swapping motor type and growing the HX.
Basis Capacity Approach Real power (2 motors) MCA
Legacy CDU (ground zero) 1.2 MW 5.0K 42.0 kW (at 120% field load) 63A
Cool-E (both generations) 2.0 MW 2.4K / 7.0K 18.0 / 18.5 kW 27 / 37A
57% / 41%
MCA reduction vs. ground zero — 57% DC-native (63→27A, 42→18kW), 41% AC-retrofit (63→37A) — both real, different bus configurations

Cool-E Spec locked pumps

Pumps: 2+1 PM motors, 238 GPM/motor ×2 = 476 GPM total , 9kW/motor cap — identical in both generations. Capacity target: 2 MW . MCA depends on bus configuration (toggle above): 27A DC-native or 37A AC-retrofit (power supply front-end added for conventional AC facilities).

How This Works

Step Logic Result
1. Facility flow vs. IT need Q = flow × ΔT on both sides. GB200's wide 20K spread means the IT side needs less flow than the facility side already provides at 476 GPM Cr = 0.827 — facility side has slack
2. Spend the slack on area, not current Growing the HX (not the pump) closes the gap to a tight approach 93.0 m² HX → 2.4K approach
3. Current follows flow only MCA is a function of pump duty (238 GPM/motor), not HX size 27A — unrelated to the 2.4K approach achieved
Why this beats a flow-scaling competitor OCP's real 157A pays for a much bigger circuit to hit the same 3°C approach, because their lever is flow, not area 83% less current, same generation

Competitor Comparison real spec

Scaling, 2MW to 24MW 1-12 CDUs

Chiller/climate analysis (ASHRAE zone check, annual bin-hour model, Legacy-vs-Modern comparison) removed from this view — confirmed no impact between Cool-E and OCP anywhere tested, so it no longer earns space alongside the panels above that show the real differentiator. Underlying methodology and data are retained in this file's source for reuse if a specific climate or legacy-IT-hardware question comes up again.