The binding constraint on AI at scale isn't the chips, it's keeping the chips cool with less energy
AI demand is exploding. Power is the bottleneck. Every megawatt used for cooling is one that can't power GPUs. Cool-E reducesthe amperage required to cool AI infrastructure, returning that capacity to compute. Built on permanent-magnet motor technology, it's ready for the higher flow demands of GB200, Rubin, and what's next.
Cool-E™ Modular Cooling System (MCS) is a patented power-agnostic, DC-native architecture that makes third-party CDUs more versatile and adaptable. Plug in on AC today, switch to DC when the facility is ready — no change to the CDU.
800V DC-Native
NVIDIA GB200 Ready
ASHRAE TC 9.9 A3/A4
NVIDIA Rubin Ready
MCA Advantage Compounds- GB200 today. Rubin tomorrow. Same architecture.
Rubin roughly doubles required flow. Cool-E answers with an added motor branch instead of pushing amperage harder while competitors scale flow through the same motors, so their amps climb faster.
Reallocated to IT per 100 MW facility, GB200-class
3+ MW
Extra racks unlocked by Cool-E’s lower MCA, GB200-class
+27 racks
Reallocated to IT per 100 MW facility, Rubin-class
7+ MW
Reallocated to IT per 100 MW facility, Rubin-class
+35 racks
Capacity Tiers — maximum continuous amperage (MCA) vs. industry baseline
IN-ROW OR ROW-END
2.4 MW
ΔT target
15–20°F
VFD losses
0%
Maximum Continuous Amperage (MCA)
Orbis Cool-E
36–37A
Copper reduction
−70 to 80%
POD
4.8 MW
ΔT target
15–20°F
VFD losses
0%
Maximum Continuous Amperage (MCA)
Orbis Cool-E
75–76A
Copper reduction
−80 to 85%
Conventional CDUs hide power loss in cooling, Cool-E reclaims that power for compute
Conventional AC CDUs rely on VFD‑driven induction pumps that quietly burn 3–5% of facility power, with losses dispersed across the plant and often invisible in audits. Cool‑E replaces that stack with native DC HaloDrive™ motors, eliminating VFD conversion loss and collapsing drive inefficiency into a single ultra‑high‑efficiency stage. At 100 MW, the recovered power is enough to run 34% more GB200‑class racks without increasing the power budget.
Dramatically lighter upstream electrical infrastructure By cutting MCA roughly 60% at the plant level, Cool‑E MCS enables much smaller conductors, panels, and switchgear than a 480 V AC, VFD‑heavy baseline. Wire gauge can step down for a like‑for‑like copper reduction approaching 80% in line runs, with correspondingly lower I²R losses and thermal loading. With no VFDs and no climate‑driven amp escalation, electrical design can be sized for steady‑state current only, simplifying coordination and compressing construction timelines.
Power‑agnostic — no site prerequisite Cool‑E MCS accepts 3‑phase 340–528 VAC or 500–850 VDC natively, so it drops into today’s AC plants and tomorrow’s DC backbones with the same hardware. Facilities can start on AC, then migrate to DC distribution when ready without touching the CDU, avoiding AC→DC rectifier losses and preventing cooling assets from becoming stranded during power‑system transitions.
Optimized ΔT cuts pump energy per MW Magnetic‑drive hydraulics and tight thermal control support a 15–20 °F ΔT where conventional induction‑pump systems typically run 10–15 °F. Higher ΔT means more heat rejection per gallon of coolant, fewer gallons per MW, and meaningfully lower pump kW for the same cooling duty. The result is higher plant COP and more of the facility’s nameplate power available for compute.
Configurable CDU architecture, consistent efficiency Whether deployed in‑row, row‑end, or at pod/plant scale, the Cool‑E Modular Cooling System keeps the same DC‑native architecture and efficiency gains. Standardized modules accelerate design and installation, and once online, energy that would have disappeared into drives, losses, and oversized copper is continuously reallocated to IT where it directly translates into more GPUs at full utilization.
White Paper: Unlocking Stranded Power Inside the Cooling Plant