xAI Co-Founder's Startup River AI Lands $1.1B to Rebuild AI Training From Scratch

River AI didn’t just raise money. It raised a flag. The startup, founded by xAI co-founder and DeepMind/OpenAI veteran Igor Babuschkin, just closed a $1.1 billion seed and Series A round led by General Catalyst and AMP PBC, with Nvidia, AMD Ventures, Y Combinator, and Temasek also piling in. For a company that only came out of stealth in June, the number is staggering — and a clear signal that some very deep pockets are betting the future of AI doesn’t belong to a handful of closed model providers.

The bet hinges on a fundamental rewiring of how AI models are built. Babuschkin’s vision, laid out in his launch blog, rejects the industry’s current obsession with creating autonomous worker replacements. Instead, River wants to make agents that are personally trainable — “guardian angels” that know you intimately and operate on your behalf. The technical path to get there means rebuilding the entire stack: training methods, model architecture, the product layer, and even new hardware designed to run AI locally. It’s an ambitious scope that makes most foundation model companies look narrowly focused.

River’s first product is already live and pointedly practical. The company offers an API that bills per million tokens and lets developers use both reinforcement learning (RL) and low-rank adaptation (LoRA) fine-tuning on open models. The pitch is a direct shot at the tyranny of prompt engineering. “Prompting steers a model you don’t own and can’t improve,” the company’s literature states. “River lets you train open models into ones that are truly yours.” Babuschkin claims an enterprise can complete a complex RL run in 15 to 20 minutes with no dedicated infrastructure team, at two to four times the cost of closed-source alternatives.

The timing is impeccable. Enterprises are increasingly queasy about vendor lock-in and are exploring multi-model strategies that include open-weight options. River is positioning itself as the post-training expertise layer that solves the hard part of that equation. And the broader vision — everyone having locally running, self-trained agents — is already bubbling up in projects like OpenClaw, while Nvidia’s partnerships with Dell, Microsoft, and HP are laying the hardware groundwork for on-device AI. What remains genuinely unclear is how River’s tech will actually differ from existing fine-tuning approaches once you look under the hood. But with $1.1 billion to spend, they’ve bought themselves a very long runway to figure it out.