Modal Labs — AI supply-chain exposure

Modal Labs runs a serverless GPU platform: AI teams push Python code and Modal handles the provisioning, scaling and teardown of the underlying accelerator clusters. It sits in the infrastructure layer as an inference-serving producer, the connective tissue between raw compute rented from the clouds and the applications that consume it, used for inference, fine-tuning and batch jobs.

Its structural hook is abstraction rather than hardware. Modal owns none of the silicon; its leverage comes from making GPU capacity feel like an on-demand function, eliminating the cluster management that otherwise gates smaller AI teams. That places it downstream of accelerator supply, exposed to the same scarcity everyone else is, but with pricing power anchored in developer lock-in and cold-start performance rather than in owning the underlying machines. The model reads its centrality as moderate: one of several serving layers, valuable but substitutable.