Cerebrium

Cerebrium is a serverless AI infrastructure platform for deploying machine learning models to GPUs. It supports 10+ GPU types including T4, A10, A100, H100, and H200, with per-second billing so you only pay for actual inference time. Models auto-scale to handle 10K+ requests per minute with sub-5 second cold starts. Deploy using standard Python code with no migration needed, with built-in support for batching, websockets, and ASGI apps. Backed by Y Combinator, used by Tavus, CivitAI, and Twilio.