Baseten recently raised a funding round backed by NVDIA to advance it’s inference stack.

Baseten has seen 100x growth in inference volume. As AI experimentation ends, companies are now focused on the high volume day to day costs and technical hurdles of keeping models running for millions of users.

Baseten’s thesis is that the future will be dominated by specialized models tailored to specific tasks. Their growth is due to recent inflection points in the open source community. Models like Llama, Deepseek and Qwen have allowed companies to move away from closed APIs and run their own custom models. The emergence of Whisper(audio) and Stable Diffusion(Image), created a demand for infrastructure that could handle multi-modalities beyond text. Also, modern teams are using reinforcement learning to fine tune models directly for specific uses, necessitating a platform that can handle these custom deployments.

As AI models transition towards “reasoning” models, computational costs increases by orders of magnitude. Baseten is positioning itself as the essential link between AI potential and real world impact. Baseten will be focusing on three pillars, performance, reliability and developer experience.

Baseten views inference as one of the largest market opportunities ever created. With the new capital they intend to expand from a depolyment tool to a full scale AI infrastructure platform. Their goal is to capture the increasing demand for compute heavy reasoning and as AI is adopted more and more Baseten becomes the agent that powers it.