Beam

Beam provides high-performance cloud computing for AI agents, developers, and teams building AI applications. Its platform combines fast-booting, autoscaling compute with tools for running code, serving models, and handling large workloads. Products include secure, persistent microVM sandboxes that can be paused, resumed, forked, and GPU-accelerated; serverless inference endpoints with OpenAI-compatible APIs; and task queues for distributing jobs across containers.

Beam supports use cases such as open-source language model inference, reinforcement learning, LoRA and QLoRA fine-tuning, ComfyUI workflows, GPU training, batch processing, and image generation. Customers can use a range of GPUs, including B200, H200, H100, A100, and RTX models, and bring their own images, volumes, and snapshots. Features include usage-based compute and storage, automatic scale-to-zero, persistent VM addresses, network controls, and secret handling through an egress proxy. Beam also offers an SDK, command-line tools, MCP tools for agent-driven cloud operations, documentation, pricing information, technical resources, and a developer community. The company describes its service as a bare-metal cloud designed for agents and is backed by Y Combinator.