$1.1 Billion to Train and Customize Your Own AI Models

River AI has raised $1.1 billion to build infrastructure that lets companies train, customize, and operate their own AI models.

River AI has raised a total of $1.1 billion across its Seed and Series A rounds. General Catalyst and AMP PBC led the financing, with strategic investments from NVIDIA and AMD Ventures and participation from Y Combinator and Temasek.

Founded by Igor Babuschkin, a former xAI co-founder who previously worked at Google DeepMind and OpenAI, River is initially focused on infrastructure that allows companies to train, fine-tune, and operate their own models using frontier open-weight models as a foundation.

River API combines LoRA fine-tuning and reinforcement learning with automated management of compute, weight transfers, and the transition from training to inference. The company says a complex reinforcement learning run can be completed in 15 to 20 minutes at a cost two to four times lower than some proprietary alternatives. Billing is based on the tokens actually used rather than on continuously reserved GPU capacity.

This infrastructure is the first part of a broader project centered on personal AI that can learn from its user while remaining under their control. River plans to work across the full stack, from training infrastructure to personalized products and hardware that could eventually run this AI closer to the user.

In an interview with The New York Times, Babuschkin also discussed the possibility of placing a dedicated computer directly in people’s homes, allowing them to use and modify their AI without constantly depending on an external provider’s infrastructure. River also says it plans to release some of its technologies as open-source software.