River AI has raised $1.1 billion in a combined seed and Series A round led by General Catalyst and AMP PBC, two months after the startup came out of stealth in June. Founder Igor Babuschkin, a co-founder of xAI with prior stints at DeepMind and OpenAI, is building River as an end-to-end rebuild of the AI training stack, aimed at turning agents into user-trained assistants rather than worker replacements.
Nvidia, AMD Ventures, Y Combinator, and Temasek joined the round. AMP PBC, the co-lead, is an AI-focused investment firm founded in 2026 by former Andreessen Horowitz general partner Anjney Midha, whose a16z portfolio included Black Forest Labs, Mistral AI, LMArena, and OpenRouter. The financing size — $1.1 billion for a two-month-old company — is one of the largest early-stage AI rounds on record.
River's first product is a training API billed per million tokens, with rates that vary by the open-weight model used. Developers can apply both reinforcement learning and low-rank adaptation fine-tuning through the endpoint, and the resulting models are served like any other API. The company positions the offering as a replacement for prompt engineering, which it argues can only steer a model an enterprise doesn't own.
Key facts
- 01River AI raised $1.1B in a combined seed/Series A round led by General Catalyst and AMP PBC, two months after emerging from stealth in June.
- 02Nvidia, AMD Ventures, Y Combinator, and Temasek joined the round alongside AMP PBC, the new firm founded by ex-Andreessen Horowitz partner Anjney Midha.
- 03River claims enterprises can complete a reinforcement learning run in 15 to 20 minutes at 2 to 4 times the cost savings versus closed-source alternatives.
- 04Founder Igor Babuschkin co-founded xAI and previously worked at DeepMind and OpenAI.
- 05River's API supports both reinforcement learning and LoRA fine-tuning on open models, billed per million tokens.
The pitch is aimed squarely at enterprises that want to control their model choices rather than route everything through a single closed frontier lab. River claims its infrastructure lets any customer complete a complex reinforcement learning run in 15 to 20 minutes without an infrastructure team, at two to four times the cost savings versus closed-source alternatives.
That framing lands at a moment when enterprise buyers are increasingly mixing open-weight and closed models rather than committing to one vendor. The post-training layer — where a generic open model becomes a company-specific asset — is one of the least-served parts of that stack, and River is positioning itself as a neocloud purpose-built for it.
Babuschkin's broader vision goes further than enterprise fine-tuning. He argues the current architecture of AI assistants, where users call a shared model to complete tasks, is the wrong endpoint. In his launch blog, he wrote that the stack has to be rebuilt end to end — training, models, the product layer, and new hardware that lets personal AI live close to the user.
The personal-agent thesis has some early corroboration in the market. Locally-running agents built on OpenClaw and its derivatives are gaining traction, and Nvidia has been partnering with Dell, Microsoft, and HP on AI-capable PC hardware that could host personal models on-device. River has not detailed which hardware layer it plans to build, only that it intends to build one.
The gap between River's ambition and its shipped product is wide. The company has one API, an open-model fine-tuning service, and a founding team. Rebuilding training infrastructure, model architectures, a consumer product layer, and new hardware is a multi-year program that has bankrupted better-funded efforts. Enterprise fine-tuning is also a crowded segment, with hyperscalers, existing neoclouds, and every frontier lab offering their own post-training tooling.
River's counter is that it is not trying to compete on any single layer — it is trying to sell a coherent stack aimed at customers who want to own their models rather than rent them. The 15-to-20-minute reinforcement learning claim is the first concrete benchmark it has offered, and it will be tested quickly once paying customers run production workloads through the API.
The scale of this round says as much about the investor climate as it does about River. A $1.1 billion check into a two-month-old company with one API is a bet on the founder, the thesis, and the infrastructure moat — not on current revenue. The more interesting question for the AI market is whether enterprises actually shift meaningful training budget toward specialized neoclouds like River, or whether the hyperscalers close the post-training gap before the thesis has room to compound. River now has the capital to find out, and roughly two years of runway to prove the personal-agent vision is more than a pitch deck.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
