Baseten

Baseten is a San Francisco inference platform that sits on top of GPU capacity rather than owning it. Founded in 2019 by ex-Gumroad machine-learning engineers Tuhin Srivastava, Amir Haghighat, and Pankaj Gupta, the company packages open-source and customer-owned models behind a low-latency serving layer, its Truss packaging framework, and the Chains orchestration runtime for compound AI pipelines. In the neocloud landscape it is positioned as an inference-focused mid-market and long-tail platform, not a hyperscale training host: workloads run across the company's Inference Cloud on partner GPU capacity in AWS, GCP and dedicated colocation, with Baseten selling engineering support, autoscaling and cost optimization on top.

The company has become one of the more visible pure-play inference vendors thanks to a run of AI-native design-partner customers including Abridge, Cursor, Notion, Writer, Descript, Sourcegraph, Clay, Zed and Gamma. Baseten does not publish a fixed hourly GPU price sheet in the CoreWeave or Lambda style; pricing is a per-model, per-token or dedicated-deployment quote, which is part of how it differentiates against raw GPU-hour resellers.

Baseten closed a $150 million Series D led by BOND in September 2025, with CapitalG, Conviction, IVP, Spark Capital, Greylock, 01A, Premji Invest, BoxGroup and Scribble Ventures participating, and Jay Simons joining the board. That round followed a $75 million Series C led by IVP and Spark Capital in February 2025. The company remains privately held.