River AI has raised $1.1 billion just two months after leaving stealth.
The Palo Alto startup wants firms to train, own and run custom AI models.
NVIDIA and AMD Ventures joined as strategic backers alongside the lead investors.
Igor Babuschkin left xAI in August 2025. By April 2026, he had incorporated a new company in Nevada. By June, River AI had emerged from stealth. By August, it had raised $1.1 billion in one of the largest seed and Series A rounds closed by a two-month-old startup this year.
The round was led by General Catalyst, whose AI and deep-tech portfolio spans Anthropic, Anduril, and Skild AI, alongside AMP PBC, the investment firm founded this year by former Andreessen Horowitz general partner Anjney Midha.
NVIDIA and AMD Ventures joined as strategic investors, with Y Combinator and Temasek rounding out the syndicate. River declined to disclose a post-money valuation, though earlier reports had pointed to talks around a $5 billion figure before the round closed at its eventual size.
“The way AI is built today is not how it will be built in the future. 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. We started River to allow people and companies to own their intelligence,” said Babuschkin, River’s CEO and co-founder.
The global generative AI market is projected to grow from $161 billion in 2026 to more than $1.26 trillion by 2034, a 29.3% compound annual growth rate, according to Fortune Business Insights. Investors clearly think there’s room in that number for more than one winner.
A résumé built at the frontier labs
Babuschkin’s résumé spans generative modelling and reinforcement learning research at Google DeepMind, large-scale training at OpenAI, and co-founding xAI with Elon Musk in 2023.
His River co-founders bring backgrounds in xAI and Tesla across deep learning, reinforcement learning, and AI infrastructure. The company, founded in 2026, works by offering an API that runs LoRA fine-tuning and reinforcement learning on top of existing open-weight models, while handling the underlying infrastructure work that would otherwise need a dedicated engineering team.
River said enterprises can complete a full reinforcement-learning training run in fifteen to twenty minutes, at two to four times lower cost than closed-source alternatives, paying per token used rather than for idle GPU capacity.
The layer beneath the model
River isn’t trying to beat OpenAI or Anthropic at building bigger models. It’s chasing the layer underneath them.
Together AI raised $800 million at an $8.3 billion valuation in July, led by Aramco Ventures, with annual bookings past $1.15 billion. Featherless.ai closed a $20 million Series A in April, backed by AMD Ventures and Airbus Ventures, making serverless access to open models cheaper.
In China, Moonshot AI reached a $20 billion valuation after raising $2 billion, chasing the same open-weight momentum on a much larger scale. River’s pitch is broader than any of them: not just hosting or serving other companies’ models, but owning the training, tuning, deployment and, eventually, the hardware underneath personal AI.
That’s a much bigger vision than infrastructure-as-a-service, and a much harder one to pull off.
Hemant Taneja, chief executive of General Catalyst, put the round in geopolitical terms: “American leadership in AI urgently requires leadership in open weight models, while maintaining a lead in closed frontier models. Igor and the River AI team have the experience to make this happen, and we view their agenda as a priority for American resilience.”
Marc Bhargava, managing director at General Catalyst, framed it as a market gap: “There is a gap between what AI can do and what most companies actually experience. Until now, companies have lacked a cost-efficient way to train, tune, and own custom AI models. River closes this gap, helping any company build models on their own data, tailored to how they actually work.”
River plans to invest $1.1 billion in training infrastructure, personalisation software and, eventually, dedicated hardware that runs personal AI closer to the people using it.
Babuschkin has the pedigree, and now the capital, to build almost anything. Whether “personal, owned AI” is something enterprises will actually pay for at scale is a different question, and it’s the one River still has to answer.