River AI, an AI startup founded by xAI co-founder Igor Babuschkin, has secured $1.1 billion in seed/Series A funding led by General Catalyst and AMP PBC with participation from Nvidia, AMD Ventures, Y Combinator, and Temasek.
(AMP PBC is an AI-focused investment firm founded in 2026 by former Andreessen Horowitz general partner Anjney Midha, a backer of a16z companies such as Black Forest Labs, Mistral AI, LMArena, and OpenRouter.)
River came out of stealth in June with a fascinating mission. Babuschkin’s resume includes AI roles at DeepMind and OpenAI, and he intends to reinvent AI from the ground up, starting with how models are trained. This is to turn agents into personally trainable assistants, rather than replacing human workers as other AI labs are following.
“To get there, we believe we need to rebuild our stack end-to-end: training, models, product layers, and new hardware that allows personal AI to live closer to you,” he said in the launch blog.
“A competent agent will become a part of your everyday life,” he envisions, “not like the assistant you turn to today when you need a job done, but more like a guardian angel. They’ll be quietly by your side, helping you with the things that really matter to you. They’ll know you well, and they’ll be yours, not someone else’s.”
River already offers an API and charges per million tokens depending on the open model used. The API allows developers to use both reinforcement learning (RL) and low-rank adaptation (LoRA) fine-tuning on their models. This first product is intended to be an antidote to accelerate engineering. “Prompting directs you to a model that you don’t own and can’t improve. With River, you can train open models to be truly yours and serve them like any other endpoint,” the product documentation says.
While this round is an eye-popping investment for a startup and perhaps another sign of the heated AI atmosphere, River’s premise comes at a particularly auspicious time. Companies are awakening to the desire to control the fate of their AI models by using them in combination with models that include indifference weights. River promises to solve some of the post-training expertise problems with its NeoCloud product.
“Any company can complete complex reinforcement learning runs in 15 to 20 minutes, without the need for an infrastructure team, and at a cost savings of 2 to 4 times compared to closed-source alternatives,” the company said in the funding announcement.
The company’s larger vision is that everyone will have their own trained agent working for them. We are already seeing this concept in action with the rise of locally running personal agents in the form of OpenClaw and its variants. Additionally, we’ve already seen Nvidia partner with PC manufacturers like Dell, Microsoft, and HP on AI-enabled hardware.
It remains to be seen how River’s technology will differ. But it starts with a war chest full of cash to take on the challenge.
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
