Chinese AI models now account for a strikingly large share of the enterprise work flowing through US developer platforms, according to a CNBC investigation published earlier this month. On some gateways, open-weight models from Chinese labs handled between 30 and 46 percent of enterprise API token usage — a share that has stayed above 30 percent every week since early February, against a trailing-12-month average of roughly 11 percent and lows near 4.5 percent in the first half of 2025.
What is driving the shift
The through-line is cost. Open-weight Chinese models are estimated to run 60 to 90 percent cheaper than leading US frontier models, and adoption has been fast where price and capability meet. Z.ai’s GLM-5.2 saw the quickest uptake of any model tracked by one major gateway, with daily token volume climbing roughly 27-fold and customer counts around 80-fold in its first full week. As usage-based pricing makes AI budgets harder to predict, more companies are routing routine drafting and analysis to cheaper, smaller models while reserving premium models for complex work such as coding.
OpenRouter’s Justin Summerville put the gap at “60 to 90 percent cheaper.” — per CNBC
The enterprise caveats
For business leaders, the cost case comes with governance strings. Major Chinese models carry hard-coded refusals on politically sensitive topics, which can break workflows that touch those subjects, and API calls generally route through Chinese-jurisdiction servers unless teams use an intermediary gateway or self-host the weights. Security-sensitive sectors may find those constraints disqualifying regardless of price. The broader signal is that enterprise AI competition is moving from raw model capability toward cost discipline and implementation — and that the decision of which model handles which task is quietly becoming one of the more consequential calls on the tech budget.
