QCT QuantaGrid D75F-9U (8× B200, 1,440 GB)

Hyperscale-grade 9U rackmount AI server featuring eight NVIDIA B200 GPUs on an HGX B200 baseboard, equipped with 1.4 TB of HBM3e memory at 64 TB/s aggregate bandwidth, dual Intel Xeon 6 processors, 32 DDR5 DIMM slots, front hot-swap NVMe bays, and delivering up to 72 PFLOPS of dense FP4 AI performance.

Designed to train and serve trillion-parameter-class models on a single node, with fifth-generation NVLink and NVSwitch connecting all eight GPUs at 1.8 TB/s, one-to-one GPU-to-NIC topology for 400G or 800G InfiniBand and Ethernet scale-out, and full NVIDIA AI Enterprise, CUDA, and Triton support.

Air-cooled in a standard rack with roughly 10 kW of continuous draw on a three-phase feed, built by the ODM that manufactures much of the world's hyperscale fleet, and able to run a 671-billion-parameter model at FP8 or a 1-trillion-parameter model at FP4 entirely in GPU memory.

Price available on request.