American Companies Are Quietly Starting to Buy Chinese Open-Weight AI

American enterprises have begun looking beyond American labs for their models. As Chinese open-weight systems close the capability gap with closed US competitors, more companies are willing to run them — and the spending data has started to show it.

Ramp's latest AI Index, which tracks token and subscription spend across its customer base, puts the share of AI-spending businesses paying for model-serving platforms — the route to open-source and Chinese-developed models — at 6.1% in July, up from 4.5% in January.

Why buyers are looking

The attraction is not ideological. Open-weight models are cheaper, can be fine-tuned freely, and can run inside a company's own infrastructure. That last point carries the most commercial weight: it removes the worry that proprietary data is training a system a competitor will later rent.

Anthropic and OpenAI still dominate how American businesses actually buy AI. The shift described here is minor in absolute terms. What makes it notable is direction and constituency — these are procurement decisions at ordinary companies, not experiments in research labs.

The policy collision ahead

A measurable share of US enterprise AI workloads running on Chinese-developed weights is a fact that Washington's technology policy has not yet priced in. Export controls have been built around chips and cloud access, on the assumption that capability flows one way.

Open weights do not respect that assumption. A model published for download cannot be un-published, and once a Kansas insurer is fine-tuning it on its own servers, there is no shipment to intercept.