# Docker Offload Docker Offload is a managed cloud service that offloads building and running containers to the cloud using the Docker tools you already know. For [[Docker AI]], the relevant angle is GPU: Offload provides ephemeral NVIDIA L4 Tensor Core GPUs (23 GB) as cloud runners, invoked from local `docker` / `buildx` commands with port forwarding and bind mounts, so the local dev loop is preserved while inference or training runs on cloud GPUs. ## Key points - Cloud execution target for [[Docker Model Runner]] and Compose-based agentic apps when local hardware isn't enough - Pricing (Beta): 300 free GPU minutes, then $0.015 per GPU-minute - Status via `docker offload status`; metrics in Docker Desktop under Offload > Insights ## References - https://docs.docker.com/offload/ - https://docs.docker.com/offload/about/ - https://docs.docker.com/offload/quickstart/ ## Related - [[Docker AI]] - [[Docker]] - [[Docker Model Runner]]