River AI raises $1.1 billion to build AI you train yourself, not AI that replaces you

Key points

• River AI, founded in 2025 by Igor Babuschkin, raised $1.1 billion in a seed and Series A round led by General Catalyst and AMP PBC.
• Nvidia, AMD Ventures, Temasek, and Y Combinator also participated in the round.
• River's first product lets developers and businesses fine-tune, or custom-train, open AI models through an API without needing a dedicated infrastructure team.
• The company claims enterprises can complete a complex AI training run in 15 to 20 minutes at two to four times the cost savings compared to closed, proprietary alternatives.
• River's longer-term goal is AI agents, software that carries out multi-step tasks on a person's behalf, that individuals train and own.

A startup barely two months old just raised $1.1 billion. River AI, founded by Igor Babuschkin, one of the co-founders of Elon Musk's xAI lab, came out of stealth in June 2026 arguing that the AI industry is heading in the wrong direction. Most labs are building AI to replace workers. River wants to build AI that belongs to you.

Babuschkin's CV spans senior roles at DeepMind and OpenAI before he helped found xAI. That pedigree, combined with General Catalyst and AMP PBC co-leading the round, pulled in Nvidia, AMD Ventures, Temasek, and Y Combinator as additional backers.

AMP PBC is an investment firm founded in 2026 by Anjney Midha, a former general partner at Andreessen Horowitz, the prominent Silicon Valley venture firm. Midha previously backed Mistral AI and Black Forest Labs. Our story on Mistral from 11 August covers why European AI ownership is becoming a competitive argument, a framing that sits neatly alongside River's pitch about who controls a trained model.

What does River actually do right now?

River's first product is a service, priced per one million tokens (a token is roughly a word or word fragment the AI processes), that lets developers and businesses train open AI models, meaning models whose underlying code is publicly available, to suit their specific needs.

The two training methods on offer are reinforcement learning, where the model learns by trial and feedback rather than memorising examples, and LoRA fine-tuning, a technique that adjusts a model's behaviour without rewriting it from scratch. Neither requires the customer to hire a team of AI infrastructure engineers. River handles that layer.

The pitch: stop trying to coax a model you don't own into doing what you want through careful prompting. Train one that's genuinely yours instead.

For businesses, that means a complex training run in 15 to 20 minutes, with cost savings of two to four times compared to closed, proprietary AI services, according to River's own figures. Those numbers are self-reported and haven't been independently verified.

What is the bigger vision?

Babuschkin describes the endgame as AI agents, software that can plan and carry out tasks over many steps, that feel less like productivity tools and more like personal advocates. In his words: "quietly present, on your side, helping with what actually matters to you."

That vision fits a real shift in enterprise technology. Large companies are increasingly mixing open and proprietary AI models rather than locking into a single provider, and the hard part is often adapting those models after training. River's positioning itself as the specialist that solves exactly that problem.

Whether a two-month-old company can deliver at that scale is the open question. The $1.1 billion gives River a long runway to try.

What does this mean for ordinary people?

Nothing to act on today. River's current product targets developers and businesses, not individual consumers. But if personally trainable AI agents catch on, the downstream effect would be assistants that learn your habits and preferences over time without that data belonging to a big tech company. That's the promise. Delivery is years away, at minimum.