River AI, a company that emerged from stealth in June, has closed a round of $1.1B that its founders say will fund a wholesale rethinking of how AI models are built and trained. General Catalyst and AMP PBC led the investment, with Nvidia, AMD Ventures, Y Combinator, and Temasek joining.
The startup is led by Igor Babuschkin, who helped found xAI after stints at DeepMind and OpenAI. His vision inverts the industry’s current direction: instead of models designed as worker replacements, River wants personal assistants that owners can train themselves, living close to the people they serve. He has described them as guardian angels that know their users well and belong to no one else.
River’s first product is a token-billed API for fine-tuning open models with reinforcement learning and LoRA. The company argues that prompting only steers a model someone else controls, while its approach lets enterprises turn open-weight models into their own. It claims a complex RL run can complete in 15 to 20 minutes without an infrastructure team, at a fraction of the cost of closed-source alternatives.
The raise lands as enterprises grow wary of depending on any single lab and seek open-weight flexibility. Whether River’s post-training bet pays off remains to be seen, but it now has an unusually deep war chest to find out.
