River AI

Babuschkin helped stand up xAI from nothing in 2023 and built the Colossus supercluster that trains Grok. His new company inverts the frontier-lab thesis: not one giant model everyone rents, but a personal intelligence you own end to end — the hardware it runs on, the data it learns from, and the weights themselves. General Catalyst is reportedly leading talks to put up to a billion dollars behind that pitch, at a valuation near five billion, before there was a product to value.

A reported billion dollars at a five-billion valuation is chasing 'AI you own' — and the only thing you can buy today is a cloud rental you don't.

There is now a product, and it is narrower than the pitch. River's live v0.1 API lets you fine-tune open-weight models from 35 billion up to a trillion parameters — Qwen, Kimi, GLM, all Chinese releases — through a Python client, billed per token from a fraction of a cent up to about $25 per million for the largest training runs. That is a real, usable service. It is also close to a commodity: Together, Fireworks and others already rent the same thing, and the 'own your weights' crowd already runs open models locally with Ollama.

The gap is the whole story. Everything that would make this AI actually yours — local hardware in your home, models that keep learning on your data, weights that never leave your control — is a paragraph on the landing page, waiting on advances River says it hasn't made yet. What ships today is a cloud rental of someone else's open models. The sovereignty is the promise; the fine-tuning bill is the product.

The bet worth watching is not the API but the packaging: whether a founder with Colossus on his résumé can turn 'AI you own' into consumer hardware before the pitch outruns the roadmap.