Serverless compute betting that AI workloads stay heterogeneous enough to need it

Altis View

Modal is both a beta bet that AI compute demand keeps growing and an alpha bet that AI workloads remain heterogeneous enough that the winning abstraction is not a single-purpose inference API or raw GPU cloud, but a developer-friendly execution layer spanning inference, batch, sandboxes, training, and agentic workflows. The case is strongest if agentic workflows also drive durable demand for sandboxes, making Modal relevant to a new category of secure AI execution infrastructure. The key risk is whether Modal remains differentiated as compute access eases and hyperscalers improve developer tooling. Therefore a central question is whether hyperscalers choose to compete directly, or whether they view Modal as a valuable customer / partner that aggregates long-tail, spiky AI demand and turns it into underlying compute revenue.