On October 30, Nomura published a research note saying that, after recent exchanges with overseas AI infrastructure company Together AI, it believes China's open-weight models are accelerating adoption in overseas markets on the strength of better cost and capability, Sina reported. Lightweight models such as DeepSeek 4.1 Flash have become among the most popular models on Together AI's platform — even surpassing some open-weight flagship models — because their ability suffices for everyday user tasks while inference cost is markedly lower than large flagships.
Nomura noted experts see overseas model usage shifting from traditional conversational apps to agentic workflows, which involve longer context and multi-round reasoning and consume significantly more tokens per task, supporting continued inference demand. Rising interest in personal AI agents is also broadening open-weight model use from coding to wider productivity and automation scenarios. While serverless hosting of public open-weight models carries limited immediate commercial risk, dedicated inference partnerships with Chinese model developers could be more sensitive to technology-export controls and cross-border compute limits.
The note reflects how China's efficient, open-weight models are carving out global developer mindshare through economics rather than scale alone.
Source: Sina / Nomura
