Silicon Valley Can't Quit Chinese AI Models, Despite Washington's Warnings

Per Rest of World, American developers are quietly turning to Chinese AI models — specifically Moonshot's Kimi and DeepSeek — even as U.S. policymakers push to contain China's AI industry through export restrictions and national security guidance.

The drivers, according to the report, are practical: Chinese models are cheap, fast, and getting better. Startups cite cost pressures and Chinese labs' rapid development pace as reasons to overlook the geopolitical risk. The report describes the gap between official containment policy and actual developer behavior as widening — though it does not specify the scale of adoption with concrete numbers or named companies.

The dynamic creates a structural problem for Washington. Export controls on chips limit China's ability to train future models, but they do nothing to prevent American companies from using models that already exist and are accessible via API. Hardware policy doesn't map cleanly onto software distribution.

It mirrors the early debates over Huawei network infrastructure, but the AI version moves faster, is harder to audit from the outside, and involves companies with far less regulatory experience than the telecoms that navigated those trade wars.

Read the full story at Rest of World