Modal bundles inference and execution

Modal is turning agent infrastructure into a bundled compute sale, not a single API call sale. When the same account can host the model, spin up an isolated machine for the agent to run code, keep its files on a persistent volume, and reuse the same container images, a customer avoids stitching together a model host and a separate sandbox vendor. That makes Modal more deeply embedded in the agent workflow and expands revenue from inference into execution time, storage, and evaluation jobs.1 2 3 4

• An agent often needs two different things. It needs GPUs to generate the next action, and it needs a separate computer to actually do the work, like running Python, editing files, or opening a browser. Modal now sells both pieces on one platform.1 2 4
• This is a competitive shift against specialists like E2B and Daytona. Those companies focus on the agent’s working computer, while Modal can bundle that environment with inference, training, and batch jobs for teams already running AI workloads there.2 3 4
• The business impact is already material. Modal launched Sandboxes in 2023, and sandbox revenue grew from near zero in May 2025 to $300M annualized by October 2026, showing that code execution is becoming a major infrastructure budget next to model serving.1 4

The next step is a fuller agent cloud where inference, sandboxing, evals, post-training, and multi-node GPU jobs all run in one place. If agent builders keep preferring fewer vendors and less glue code, platforms like Modal should capture a larger share of AI infrastructure spend than inference-only providers.1 3 4