Comment on: ZeroGPU

The pay-for-efficiency angle is refreshing when most platforms just bill raw GPU-hours. Curious how you handle cold starts on the serverless layer — that's usually where the "compute-efficient" promise breaks for spiky workloads.