One hardware platform. Two accelerator generations. No rebuild.
Cool-E POD is a prefabricated, dual-duty liquid cooling platform for AI data centers: 12 identical 2 MW modules per 56' platform, 24 MW per platform. Every module ships with the same radial-flux permanent magnet motors and 800V DC-native SiC inverters. The duty point shifts between GB200/H100 and Rubin chiller-less by firmware profile only. No hardware change.
Built for the platform shift you can't predict.
Cooling infrastructure gets locked in years before an operator knows which accelerator generation will fill the racks. A design sized for GB200 today can be the wrong flow profile for Rubin tomorrow, forcing a costly hardware swap mid-deployment. Cool-E POD removes that risk: the same skid, in every slot, covers both duty points. GB200/H100 and Rubin chiller-less are selected by software, not by installing different hardware.
One skid, two duty points
The same 4-unit motor-pump skid runs GB200/H100 at 2:2 (N+2 of 4) redundancy or Rubin chiller-less at 3:1 (N+1 of 4). The switch is a firmware profile, not a hardware swap.
No rebuild, no downtime risk
Twelve identical 2 MW modules per 56' platform, 24 MW per platform, delivered as a single prefabricated row. Single-end field tie-in keeps commissioning straightforward.p.
Engineered for what's next
Radial-flux PM motors and 800V DC SiC inverters give Cool-E POD the headroom for Rubin's roughly doubled flow demand, without pushing amperage harder to get there..
12 identical 2 MW modules per platform
24 MW per 56' row
Rubin chiller-less & GB200/H100 compatible
One platform, tuned two ways.
Firmware sets the duty point. The hardware never changes.
2.4 K
fixed approach temp held constant at either duty point, GB200/H100 or Rubin chiller-less.
NVIDIA’S Rubin / Chiller-less
785 GPM
total flow
NVIDIA’s GB200 / H100
45°C / 55°C
supply / return
3:1 (N+1 of 4)
Active : standby
485 GPM
total flow
~39°C / 55°C
supply / return
2:2 (N+2 of 4)
Active : standby
Dual-Duty Performance
SPECIFICATIONS
2 MW
485 GPM
~37.8 kW
GB200/H100
2:2 (N+2 of 4)
2MW
39°C / 55°C
2.4 K (fixed)
785 GPM
~30 kW
Parameter
Pod footprint
Active : standby
Total flow
Supply / return
785 GPM
Rubin Chiller-less
TECHNICAL SPECIFICATIONS
Parameter
261.7 GPM (100%)
Flow / active pump
2.4 K (fixed)
Total pump power
HX approach temp
Heat rejection
242.5 GPM (93%)
Grooved (Victaulic) or flanged
12, single row, 24 MW/platform
56”L x 8.6 W x 9”H
92"/128” with headers
Pod (2 platforms)
Radial-flux PM, 800V DC SiC inverter
Redundancy
Piping
~600A @ 800V DC
Single-end hookup, one side
CDU Footprint
N+1 (Rubin) / N+2 (GB200-H100)
24 MW /48 2 pods
CDU overall height
Motor / drive
Power tie-in
Field tie-in
2 MW modules/pods
Rubin Chiller-less
48"x 48, pump/motor skid
785 GPM
45°C / 55°C
Active : standby
785 GPM
Pod/CDU Module
24 MW in a single 56' row.
Twelve identical 2 MW CDU modules line a single platform, each built around the same 4-unit motor-pump skid. Two platforms combine for 48 MW. Every slot uses the identical hardware SKU; the duty point is set by firmware, not by which module the site installs.
12 identical 2 MW CDU modules per row • one hardware skid SKU in every slot • duty set by firmware
Smart controls keep every module at its best efficiency point.
Cool-E POD's control layer runs the same logic regardless of duty point, so switching between GB200 and Rubin never means retuning the system.
Master loop control holds a return-temperature / ΔT setpoint, using the same control law for either duty.
Per-pump SiC inverter field-oriented control balances load across the manifold automatically.
Staging hysteresis adds capacity above 90% load and sheds below 65–70%, keeping pumps near their best efficiency point.
Bumpless transfer switches active/standby roles with no interruption to cold plate flow.
Run-hour wear-leveling rotates duty across all 4 units in every skid, extending service life fleet-wide.
White Paper: Unlocking Stranded Power Inside the Cooling Plant
