NVIDIA B300 (cloud, 288 GB)

Rented NVIDIA B300 SXM GPU with 288 GB of HBM3e at 8 TB/s, native FP4 at roughly 1.5× B200 throughput, and fifth-generation NVLink at 1.8 TB/s when rented as an 8-GPU node.

Designed for bursty training and inference with no capital outlay, billed per GPU-hour from $3.67 per GPU-hour spot to $8.55 on-demand on early-access clouds, and scalable from one GPU to hundreds in minutes with CUDA, PyTorch, and the NVIDIA container stack preinstalled by every provider.

Lives in someone else's data center: nothing to power, cool, or secure, and the trade is per-hour cost and data leaving your premises. Runs a 288-billion-parameter model at 8-bit on one GPU; 1 trillion at FP8 on eight.

$3.67/hr spot, $8.55 on-demand, up to $18 on hyperscalers. Early availability; verified September 2026.