Chinese AI models have quietly become the default choice for a large share of the world's developers. According to usage data shared with CNBC, models from Chinese companies now account for the majority of token traffic on two major developer platforms that act as gateways to models from many providers.
The numbers represent a dramatic reversal in less than a year. On OpenRouter, Chinese models accounted for 57% to 67% of tokens used in the week of September 14, up from just 6% to 13% in February. On Vercel, their share rose to 55% in August from 11% in January.
For anyone following the latest AI developments, the data punctures a core assumption of the past two years of AI policy: that export controls would keep Chinese models marginal outside China. Instead, Chinese open-source models are winning developers on the two metrics that matter most in production environments — capability at the task level, and price.
The numbers behind the surge
OpenRouter's data covers companies in the US, Europe, and what the platform defines as the "Global South" — 82 countries across Central and South America, Africa, and Asia. Vercel did not specify a geographical breakdown for its figures.
Businesses in the Global South have been the biggest users of Chinese AI models on OpenRouter in recent weeks: 67% of the tokens those companies consume are on Chinese models. That matters because about half of all tokens on OpenRouter are used by companies in the US — meaning American firms still make up the largest single block of demand on the platform, even as Chinese models capture the majority share of what those tokens run on.
Cheap, good enough, and open
The shift is being driven less by any single breakthrough than by a widening price-performance gap. Peter Walker, head of insights at OpenRouter, told CNBC that Chinese open-source models released this year "can credibly perform in advanced agentic use cases, especially in regards to coding, in a way that was just not true in late 2025." They are also "incredibly cost-effective compared to most models from American labs," he said.
Harpreet Arora, head of agentic infrastructure at Vercel, put the economics bluntly: "Chinese models are becoming capable enough for more tasks at a much lower cost. Once a model meets the quality bar for the job, that price difference becomes compelling."
The caveat, Arora noted, is that companies still want frontier US models for the most complicated tasks. The pattern emerging on both platforms is a tiered one: Chinese models absorb high-volume, cost-sensitive workloads, while American frontier models are reserved for the hardest problems.
The capability gains are real, according to CNBC's reporting. Chinese companies including DeepSeek, Z.ai, and Alibaba have released new models this year with major performance gains in tasks such as coding — the workload that most developer-platform traffic consists of.
Washington is paying attention
The adoption surge has not gone unnoticed in the US capital. Two US House Committees are investigating the impact of rising Chinese model adoption, CNBC reported, and the findings could shape the next round of AI export policy.
Washington's existing toolkit is already under strain. The US has sought to preserve its AI lead by restricting Chinese AI companies from buying the most advanced chips through export controls. But officials are now concerned about two workarounds: Chinese labs accessing Nvidia chips remotely via overseas data centers, and "distillation" — the practice of training new models to mimic the outputs of older, more established ones.
Daniel Remler, a senior fellow in the technology and national security program at the Center for a New American Security, told CNBC that Chinese AI represents "real economic and security risks for the United States."
"The ultimate concern is that the integration of Chinese AI models pulls countries into a Chinese technology sphere of influence that hardens into geopolitical alignment," he said.
US labs respond on price
American labs are not standing still. Earlier this week, both OpenAI and Anthropic announced new, cheaper models — a competitive response aimed directly at the cost gap that has fueled Chinese model adoption.
Dianne Penn, head of product management, research and labs at Anthropic, told CNBC the company is trying to make its models' answers "more efficient, so it uses less tokens depending on your effort setting." Token efficiency is another way to cut effective prices without cutting list rates — and with token volume now the measuring stick for market share on routing platforms, it is a metric that will only grow in importance.
What it means
The market share data reframes the US-China AI race. The conventional narrative has focused on frontier benchmarks, where the most advanced US models still lead. But the token economics tell a different story: for a majority of real-world developer workloads on two major gateways, Chinese models are now the rational default.
That creates a policy dilemma with no obvious answer. Export controls aim to slow China's frontier progress, but the models Chinese labs ship openly — free to download, cheap to run — are spreading fastest precisely where US policy has least reach. Meanwhile, AI has become a central topic at the highest levels of diplomacy, with AI a major focus as President Trump and President Xi met in Washington this week.
The next data points to watch: whether OpenRouter's and Vercel's Chinese-model share keeps climbing through the end of the year, and whether US labs' cheaper models blunt the price advantage before adoption hardens into dependency.
---
Stay Ahead of AIThe AI market is shifting faster than ever. Get the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
