Beam is a platform offering serverless AI compute for running sandboxes, inference, and training workloads on GPUs. It enables running AI workloads across multiple clouds with sub-second cold starts, massive parallelism, and full observability. Beam provides a developer-friendly experience with no infrastructure to manage, supporting stateful sandboxes, durable task queues, and GPU inference. The platform supports bring-your-own-cloud with no lock-in and allows switching hardware easily. Beam is backed by Y Combinator.