I've been following this company for a while now, and others like it, and I'm really wondering how the economics of these rent-a-big-GPU services shake out. The specs say you get a GeForce GTX 1080 (or equivalent, so maybe they're licensing NVIDIA vGPUs or something)... do they actually have one big GPU per subscription, or do I just have access to a big GPU whenever I want? You can't slice up a GPU as effectively as you can slice up CPU cores... Filling up a datacenter with big GPUs seems like a very expensive proposition.