A year after leaving xAI, Igor Babuschkin launches River AI, advocating for personalised, user-controlled AI models and challenging the centralised dominance of major corporations in artificial intelligence.
A year after leaving xAI, Igor Babuschkin has returned to the centre of Silicon Valley’s debate over artificial intelligence ownership, this time with a start-up that argues AI should be personal, customisable and outside the control of any single company. His new venture, River AI, is promoting a model in which users can run and shape AI systems themselves rather than depend on a central platform.
The move follows Babuschkin’s departure from xAI in August 2025, when he said he was leaving to pursue a new phase of work focused on AI safety and research. Reporting at the time said he had been a key engineering figure at xAI and had previously worked at Google DeepMind and OpenAI before joining Elon Musk’s company. He later set up Babuschkin Ventures, an investment firm aimed at backing AI safety research and young companies working on the technology’s long-term potential.
River AI’s pitch is more ambitious than a conventional software service. The company says it wants to put a new kind of computer in people’s homes so they can own and use AI directly, without oversight from River AI itself. Its public materials say the business plans to release its technology as open-source software, allowing developers, companies and consumers to use and modify it. That approach places it among a growing group of advocates in Silicon Valley who argue that AI should be distributed more broadly rather than concentrated in the hands of a few large operators.
The company also says its API is designed for actual model training rather than simple prompting. According to River AI’s website, users will be able to fine-tune open-source models using LoRA, a method that adapts a model efficiently with fewer resources, and reinforcement learning, a technique that rewards desired outputs during training. River AI says the system will support models ranging from 35 billion to 1 trillion parameters, which indicates a push towards high-capacity systems that users can claim as their own.
Disclaimer: This content is intended for informational purposes only. Readers are advised to exercise their own judgement, conduct due diligence, or consult a qualified expert before acting on any information provided.
