General Catalyst leads $1.1B round into 2-month-old River AI

River AI, an AI startup founded by xAI co-founder Igor Babuschkin, has secured $1.1 cardinal successful backing successful a seed/Series A circular led by General Catalyst and AMP PBC, with information from Nvidia, AMD Ventures, Y Combinator, and Temasek. 

(AMP PBC is an AI-focused concern steadfast founded successful 2026 by erstwhile Andreessen Horowitz wide spouse Anjney Midha, backer of companies astatine a16z similar Black Forest Labs, Mistral AI, LMArena and OpenRouter.)

River came retired of stealth successful June with a fascinating mission. Babuschkin, whose resume includes AI roles astatine DeepMind and OpenAI, intends to reinvent AI from scratch, opening with however models are trained. This is successful bid to crook agents into personally trainable assistants, alternatively than pursuing the trajectory different AI labs are on: quality idiosyncratic replacements.

“To get there, we judge the stack has to beryllium rebuilt extremity to end: training, models, the merchandise layer, and caller hardware that lets idiosyncratic AI unrecorded adjacent to you,” he wrote successful his motorboat blog.

“Capable agents volition beryllium a mean portion of mundane life. Less similar the assistants you telephone connected contiguous erstwhile you request a task done, much similar guardian angels: softly present, connected your side, helping with what really matters to you. They volition cognize you well, and they volition beryllium yours, not idiosyncratic else’s,” helium envisions.

River already offers an API, billed per 1 cardinal tokens, with rates babelike connected the unfastened exemplary used. The API allows developers to usage some reinforcement learning (RL) and low-rank adaptation (LoRA) fine-tuning connected the models. This archetypal merchandise is intended to beryllium an antidote to punctual engineering. “Prompting steers a exemplary you don’t ain and can’t improve. River lets you bid unfastened models into ones that are genuinely yours — and service them similar immoderate different endpoint,” the merchandise lit says.

While this circular is an eye-popping-size concern for a nascent company, and possibly different denotation of the overheated AI atmosphere, River’s premise comes astatine a peculiarly auspicious time. Enterprises are waking up to wanting to power their AI exemplary destiny by utilizing a premix of models, including unfastened weight. River is promising to lick the post-training expertise portion of that occupation with its neocloud offering.

“Any endeavor tin implicit a analyzable reinforcement learning tally successful 15 to 20 minutes with nary infrastructure squad required, astatine 2 to 4 times the outgo savings comparative to closed-source alternatives,” it wrote successful its backing announcement.

The bigger imaginativeness of this institution is that everyone volition person their ain agents, trained by themselves, and moving connected their behalf. We are already seeing this conception successful enactment with the emergence of personal, locally-running agents successful the signifier of OpenClaw and its derivatives. Plus, we are already seeing Nvidia spouse with PC makers similar Dell, Microsoft, HP for AI-capable hardware.

How River’s tech volition disagree remains to beryllium seen. But it’s starting retired with a warfare thorax afloat of currency to try.