[GPU Math] Why Bursty AI Agents Make Terrible GPU Tenants

Last month I caught myself pricing out a dedicated A100. The logic felt airtight: my agents run around the clock, the token bill shows up every month like rent, and open-weight models are genuinely go