Modal reimagines cloud computing for the AI era by replacing traditional container orchestration with a decorator-based Python SDK that turns local functions into serverless cloud workloads. Developers define compute requirements, GPU types, container images, and storage volumes entirely in Python code rather than YAML configuration files or Dockerfiles. The platform spins up GPU-enabled containers in as little as one second with cold starts typically between two and four seconds, making it viable for latency-sensitive inference workloads that previously required dedicated GPU capacity.
The platform provides elastic access to NVIDIA GPUs ranging from T4s to H100s and B200s through partnerships with Oracle Cloud Infrastructure, with automatic scaling from zero to hundreds of concurrent containers. Modal Volumes offer a high-performance distributed file system for sharing data between function runs, while Sandboxes provide secure ephemeral environments for testing AI models and running untrusted code. The integrated Notebooks feature enables real-time collaborative development with cloud GPU access, and built-in logging provides full visibility into every function and container execution.
Modal attracted significant industry adoption with customers including Meta, which used it to run the Code World Model neural debugger across thousands of concurrent sandboxed environments, and Scale AI, which relies on it for massive evaluation spikes and MCP server orchestration. The platform raised an $87 million Series B in September 2025 at a $1.1 billion valuation. A generous free tier provides $30 in monthly compute credits, making it accessible for individual developers and prototyping before scaling to production workloads.