River AI raises $1.1B to build 'personally trainable' agents, two months out of stealth

The Palo Alto startup from former xAI co-founder Igor Babuschkin drew in General Catalyst, AMP PBC, Nvidia, AMD Ventures, Y Combinator, and Temasek on a thesis that enterprises — and eventually individuals — should own and retrain their own open-weight models.

River AI, a Palo Alto startup founded by former xAI co-founder Igor Babuschkin, said Tuesday it had raised $1.1 billion roughly two months after emerging from stealth in June. General Catalyst and Anjney Midha’s AMP PBC, itself founded in 2026 by the former Andreessen Horowitz general partner, co-led. Nvidia, AMD Ventures, Y Combinator, and Temasek joined. The company declined to disclose a valuation, per Reuters, and employs about 20 people.

The size of the round relative to the headcount is the point, not an anomaly. Capital is flowing to whichever operators can plausibly claim to rebuild the frontier stack outside the closed labs, and Babuschkin, who worked on generative modeling and reinforcement learning at Google DeepMind, led large-scale training at OpenAI, and co-founded xAI with Elon Musk before leaving roughly a year ago, is a credentialed vessel for that thesis.

River’s opening product is an API for LoRA fine-tuning and reinforcement learning on frontier open-weight models, with token-metered billing and instant deployment. The company claims a complex RL run takes 15 to 20 minutes without a dedicated infrastructure team, and pitches two to four times the cost savings versus closed-source alternatives. The longer-term vision, as Babuschkin described in his launch blog and to TechCrunch, is to rebuild “end to end: training, models, the product layer, and new hardware that lets personal AI live close to you.” Per the New York Times, that includes a server-class computer meant to keep private data off River’s own machines.

“AI should be open, freely available, and affordable. It should feel like it is working for the person using it, not the lab that trained it,” Babuschkin said. Elsewhere he framed it structurally: “We hope that in the future, AI will be trained to benefit you as the individual, rather than being a centralised thing that is controlled by one big corporation.”

General Catalyst chief executive Hemant Taneja gave the round its geopolitical read-out. “American leadership in AI urgently requires leadership in open weight models,” he said, describing Babuschkin to the Times as “one of the very few people” who could compete with China’s open-source labs. “Hopefully, this is America’s answer.”

The framing is worth noting. Open-weight AI has spent two years as a hobbyist and researcher cause. It’s now being underwritten, at billion-dollar scale, as industrial policy.

River AI raises $1.1B to build 'personally trainable' agents, two months out of stealth

Sources