#Industry News
AI Compute as a Service, the Native Liquid-Cooled Container Path
72 B300 GPUs, 1.1 Exaflops and 800 kW per Box: Full-Chain Liquid Cooling Carrying 1,400 Watts per Chip From the Moment the Container Lands
EXTRCOOL turns AI compute into a 45-foot container that deploys the NVIDIA GB300 NVL72 natively: each supernode fuses 72 B300 GPUs with 36 Grace CPUs for up to 1.1 exaflops of FP4 inference, four to six supernodes per box push IT load toward 800 kilowatts, so full-chain liquid cooling carries away up to 1,400 watts per chip through manifolds and quick couplings to every cold plate. Two integration designs fit different needs, full-rack direct-mount for native fastest deployment and modular air-liquid shared racks for user-built storage, switches and power, and two power configurations accept DC with integrated HVDC or standard AC input. The box powers on the moment it lands, on-site work cut to water and electricity hookup with commissioning compressed from months to days. EXTRCOOL provides one-stop liquid cooling solutions.
