A startup founded this year by former xAI co-founder Igor Babuschkin has collected $1.1 billion across its seed and Series A rounds, an unusually large war chest for a company arguing that AI should belong to its users rather than the labs that train it.
The round was co-led by General Catalyst and AMP PBC, with Nvidia and AMD Ventures joining as strategic investors and Y Combinator and Temasek also participating, the company said.
River AI’s debut offering is an API for fine-tuning open-weight models on a customer’s own data, removing the need to assemble the heavy infrastructure that large-scale training usually demands. Models from 35 billion up to 1 trillion parameters are supported, and complex reinforcement learning jobs can wrap up in 15 to 20 minutes, per the startup, which also claims savings of two to four times versus closed-source alternativesernatives.
Behind the scenes, River handles weight transfers, sampling-training consistency and elastic compute, then moves models straight into production with customers paying for tokens consumed during training.
The thesis is simple: as models balloon past the trillion-parameter mark, self-hosting becomes impractical, and the economics shift to whoever runs reliable inference. Babuschkin, who helped build xAI’s Grok, wants River to be that layer for open-weight models.
The round ranks among the largest ever raised by a company this young and puts River directly in competition with entrenched AI clouds. Whether enterprises trust a newcomer with their model weights remains to be seen, but the check sizes from Nvidia and AMD suggest the chip giants see value in a neutral fine-tuning layer that runs on any hardware.
