River AI’s $1.1B Round Turns Custom-Model Training Into a Co-Op Between NVIDIA and AMD

River AI, founded by former xAI co-founder Igor Babuschkin, announced it raised $1.1B. The round was led by General Catalyst and AMP PBC, with strategic investment from both NVIDIA Ventures and AMD Ventures; Y Combinator and Temasek also participated.

River describes a platform that exposes custom-model creation through an API, including reinforcement learning (RL) training and LoRA fine-tuning for frontier open-weight models. River’s pitch is explicitly workflow-driven: it claims a complex reinforcement learning training run can finish in 15 to 20 minutes, “without an infrastructure team,” and that trained models can be deployed instantly to production.

The cost story is similarly workflow-linked. River says billing is metered on tokens used for training and inference, and it claims “two to four times” cost savings versus closed-source alternatives—an argument aimed at shrinking the “GPU time + orchestration overhead” tax that has traditionally kept custom training out of reach for many enterprises.

It’s unusual for rival chip ecosystems to co-fund the same early-stage custom-model infrastructure company—especially when the company is betting on open-weight models and enterprise training workflows rather than a narrow inference-only product.

Put simply: River is trying to make custom training feel like a standard software call. If it succeeds, chip vendors can no longer treat “enterprise training” as a downstream consequence of their hardware—they must compete on orchestration layers, software compatibility, and performance reliability across customer workflows.

A workflow that claims 15–20 minute RL runs and instant deployment to production implies repeated cycles of training and inference. Even without disclosing customer deployment counts, the architecture economics point to a standard “compute stack” dependency chain: GPUs/accelerators, fast interconnect, adequate high-bandwidth memory capacity, and reliable training runtime software that doesn’t collapse under elastic provisioning.

River AI’s $1.1B raise—led by General Catalyst and AMP PBC with strategic participation from both NVIDIA and AMD Ventures—signals that enterprise demand for “custom, owned” models may soon be constrained more by training workflow tooling than by model availability. If River’s API can truly make RL fine-tuning and deployment turnkey, the next competitive battleground for chip vendors won’t be only benchmarks; it will be the ability to deliver reliable training-and-serving experiences that enterprises can repeatedly execute with minimal internal effort.