Baseten Raises $300 Million, Valuation Soars to $5 Billion in AI Funding Round

Baseten was founded in 2019, quietly, by four people who had already seen this movie before and didn’t like the ending. Tuhin Srivastava, Amir Haghighat, Philip Howes, and Pankaj Gupta had spent more than a decade building, breaking, and rebuilding machine learning systems that worked beautifully in notebooks and collapsed the moment real users showed up. San Francisco became home base, New York the second front, and the mission was simple in a way only hard problems ever are. Training models was getting cheaper and louder. Running them in the real world was still a mess.

Baseten just closed a $300M Series E at a $5B valuation. Institutional Venture Partners and CapitalG led the round. Nvidia Corporation put in $150M, exactly half the raise, which is not subtle behavior. This is the chip king walking into the server room and saying this layer matters. Inference is no longer backstage. It is the show.

Baseten lives in that moment after the model is trained, when latency, uptime, cost, and compliance decide whether AI is a feature or a liability. The team calls it AWS Lambda for AI workloads, but what it really feels like is pressure relief. Orchestration software, optimized GPU runtimes, and deployment tooling that let teams ship without inventing their own infrastructure religion. That matters when over 100 enterprises are already running mission critical workloads and thousands more developers are leaning on the platform to stay fast and sane.

The numbers tell their own story. Revenue grew more than 10x year over year by late 2025. Customers like Patreon cut GPU costs by roughly 70% and saved hundreds of developer hours annually. Healthcare platforms are hitting sub 160 millisecond latency while staying HIPAA compliant. This is production, with consequences.

Tuhin Srivastava is still CEO, still technical, still allergic to fluff. Amir Haghighat, Philip Howes, and Pankaj Gupta are still building where it hurts, deep in the runtime and the scheduler. In August, Dannie Herzberg joined as President to scale go to market muscle after doing the same at HubSpot and Slack. Jay Simons from BOND joined the board in September. Sarah Guo at Greylock has been there since the beginning. Continuity like that is a signal.

Baseten keeps its name honest. Base layer. 10/10 reliability. The part nobody claps for until it fails. With Nvidia now financially aligned and inference spending becoming the recurring bill that never stops, the floor is getting crowded and the tempo is rising. If training was the warm up, this is the long set, and Baseten is already plugged into the system, watching the meters, nudging the gain, seeing who is about to push production a little harder than last quarter.