Serverless GPU Compute

Run training, fine-tuning, and inference workloads on serverless GPUs without managing infrastructure.

**Provider:** [Modal](https://modal.com)

**Pricing:** Usage-based, from ~$0.0005 / GPU-second

**Best for:** Teams that need elastic GPU access for LLM inference, fine-tuning, or batch jobs

Modal provisions GPUs on demand and bills by the second. It supports PyTorch, Hugging Face, vLLM, and custom containers, making it a good backend for AI agents that need occasional heavy compute.