Baseten — AI supply-chain exposure

Baseten is an inference platform — the layer that takes a trained model and actually runs it in production, handling the serving, scaling and latency that turn a set of weights into a live API. Its customers are largely AI-native companies that would rather rent optimised serving than build and operate their own accelerator fleets. In the chain it is a producer of inference: it sits above the raw compute, packaging accelerators into a service that developers simply call.

Its structural position rides the shift of AI spend from training toward inference — the repeated, per-request cost of running models in the wild, which grows with every user rather than every model. That makes independent serving platforms a fast-growing but contested layer; they depend on the underlying chips and clouds and compete on optimisation and reliability rather than on a scarce input they alone hold. The chain places Baseten in its middle band: real and growing exposure to inference demand, but a position built on execution rather than an unavoidable chokepoint.