River AI has raised $1.1 billion in a combined seed and Series A round, an extraordinary sum for a company that emerged from stealth only in June. General Catalyst and AMP PBC led the round, with participation from Nvidia, AMD Ventures, Y Combinator and Temasek.
The startup is the second act for Igor Babuschkin, who co-founded xAI before departing. His background runs through the industry's most demanding training environments: generative modeling and reinforcement learning work at Google DeepMind, followed by large-scale training efforts at OpenAI.
A Different Bet on Agents
Babuschkin's thesis diverges sharply from the direction most frontier labs have taken. Where competitors are building systems pitched as replacements for human workers, he wants agents that individuals train and own themselves. Getting there, he has written, requires rebuilding the stack end to end: training, models, the product layer, and new hardware that keeps personal AI physically close to its user.
He has described the endpoint in unusually personal terms, comparing capable agents less to on-demand assistants than to guardian angels that are quietly present, aligned with the user's interests, and belong to that user rather than to a vendor. His stated goal is to shift ownership of AI away from the labs that train models and toward the people and organizations that actually use them.
The First Product
River's initial offering is already live: an API billed per million tokens, with rates that vary by the open model selected. Developers can apply both reinforcement learning and low-rank adaptation (LoRA) fine-tuning to those models, which the company positions as an alternative to prompt engineering. Its argument is blunt. Prompting steers a model you neither own nor can improve, while training produces an endpoint that is genuinely yours.
The performance claims are aggressive. River says any enterprise can finish a complex reinforcement learning run in 15 to 20 minutes without a dedicated infrastructure team, at two to four times the cost savings of closed-source alternatives.
Timing and Context
The round lands at a moment when enterprises are actively trying to reduce dependence on any single model provider. Mixed deployments that combine proprietary and open-weight models have become common, and the expertise gap tends to appear at the post-training stage, which is precisely where River is aiming its neocloud offering.
The broader shift toward personal, locally running agents is already visible elsewhere, most notably in the rise of OpenClaw and its derivatives. Hardware is moving the same direction, with Nvidia partnering with Dell, Microsoft and HP on AI-capable machines. Nvidia's presence on River's cap table sits comfortably alongside that strategy.
AMP PBC, one of the two lead investors, is itself new. The AI-focused firm was founded in 2026 by Anjney Midha, formerly a general partner at Andreessen Horowitz, where he backed Black Forest Labs, Mistral AI, LMArena and OpenRouter.
An Open Question
A $1.1 billion round for a two-month-old company will read to some as evidence of an overheated market, and River has not yet shown how its training stack differs technically from what already exists. What it does have is capital on a scale that buys years of runway and access to scarce compute.
Whether personally trained agents become a mainstream category or remain a developer niche will decide whether that bet pays off. For now, River is one of the best-funded attempts to find out.
