Nvidia backs Baseten with $300m round to scale AI inference

This deal underlines just how central inference has become in the post‑“train the giant model” phase of the AI cycle. Baseten isn’t building new foundation models; it’s selling the plumbing that gets those models into production at scale. Nvidia writing a $150 million check—half the round—is an explicit bet that controlling the inference layer will help keep demand for its chips elevated even as specialized accelerators and rival clouds crowd the market.

Strategically, Nvidia is executing a classic ecosystem play: seed the companies that are most likely to burn GPU hours, then knit them into a stack that runs best on Nvidia hardware. Baseten already supports Nvidia’s own models and hardware; deeper financial alignment increases the chance that new inference features land first and best on Nvidia platforms. For rival cloud providers and chipmakers, this raises the competitive temperature: if more of the high‑growth inference startups are partially “Nvidia‑aligned,” alternatives like AMD, Google TPUs or custom ASICs may find it harder to displace CUDA in production.