NVIDIA GB200 NVL72 Rack-Scale System

The NVIDIA GB200 NVL72 is a complete liquid-cooled rack that connects 36 Grace CPUs and 72 Blackwell GPUs through a 72-GPU NVLink domain, so the whole rack behaves as one accelerator with 13.4 TB of HBM3e. Eighteen compute trays each carry two GB200 Grace Blackwell Superchips, nine NVLink switch trays provide 130 TB/s of aggregate NVLink bandwidth, and the Grace CPUs add 2,592 Arm Neoverse V2 cores with up to 17 TB of LPDDR5X. Sparse NVFP4 throughput reaches 1,440 PFLOPS per rack, with FP8 at 720 PFLOPS, which is why NVIDIA positions it for trillion-parameter real-time inference.

GB200 NVL72 is a product for hyperscalers, national AI initiatives, large model developers and colocation operators that already have, or are building, facilities with direct-to-chip liquid cooling and 100 kW-class rack power. It is normally delivered as racks integrated by partners such as Supermicro, Dell, HPE or Foxconn, and multiple racks are joined with Quantum-X800 InfiniBand or Spectrum-X Ethernet to form larger clusters.

InTechnologies FZCO supports GB200 NVL72 projects in the UAE and surrounding regions as a sourcing and logistics partner. Rack-scale GPU systems are sold under contract and invoice from our licensed partners; InTechnologies coordinates the OEM order, consolidates crated racks and accessories in Dubai and arranges DDP or DDU delivery with suitable heavy-freight handling. Export licensing and end-user verification under US export regulations are mandatory for this product and are completed before any delivery date is confirmed.