ORBIS ELECTRIC
O|E
100 MW IT AI SCALE REFERENCE POINT
Staying Ahead of NVIDIA's
Next Chiller-less Generation
How Cool-E™ Potentially Reallocates 8+ MW to IT / Rubin-Class

Rack Reallocation

Rubin: 102 racks
freed per 100MW, deployed on Rubin chips (4 branches active), vs. a projected Rubin-era competitor
DC-native
AC-retrofit

GB200 Cool-E vs. Rubin Cool-E

Buying Rubin-ready hardware costs nothing if you only run GB200 chips today. Same current draw, plus better chiller margin as a bonus — you're just not using the 4th branch yet.

ConfigBranchesFlowApproachMCA
GB200 Cool-E2+1476 GPM2.4K27A
Rubin Cool-E3+1787 GPM2.4K49A
Rubin hardware, running GB200 only2 of 4476 GPM1.1K27A

Cool-E Rubin Spec

2.4K
Approach
42.6°C
Tci_max
787
GPM / 2MW CDU
3+1
Motors / branches
49A
MCA, DC-native / CDU
68A
MCA, AC-retrofit / CDU

How This Works

StepLogicResult
1. Same heat, tighter spreadQ = flow × ΔT. Same 2MW, but Rubin's ΔT is half GB200's (10K vs 20K)Total flow: 476 → 787 GPM
2. Split across a new motorCool-E adds a branch instead of pushing existing motors harderPer-motor flow: 238 → 262.3 GPM (+10%)
3. Current follows flow², per motor1.10² = 1.22 — a real physical relationship, not a guessPer-motor current: +22%
4. More motors, too2 active → 3 active×1.5 motor count
Total1.22 × 1.5 = 1.82MCA: 27A ‚Üí 49A (+82%)

Competitor Projection estimated

Scaling, 2MW to 24MW 1-12 CDUs