Which workloads fit an off-grid RTX GPU unit?

A builder says they're developing an off-grid compute unit with two RTX PRO 6000 GPUs per node. They want to hear from people renting RTX GPUs rather than B300 or H200 clusters, especially for workloads needing no more than 96GB of VRAM per GPU, high-bandwidth GPU links or exceptionally fast, low-latency internet. They suggest smaller-model inference, robotics simulation, fine-tuning and rendering as possible uses.