Chinese AI Models Reach Majority Token Share on OpenRouter and Vercel Developer Platforms

THE ESSENTIALS

Chinese AI models reached 57-67% token share on OpenRouter in mid-September 2026, up from 6-13% in February
On Vercel, Chinese model usage rose to 55% in August from 11% in January
OpenRouter says Chinese open-source models can now credibly handle advanced agentic tasks, especially coding
Two U.S. House committees are investigating the national security implications of rising Chinese AI adoption
Cost-effectiveness relative to U.S. frontier models is a primary driver of the adoption shift

Token Share Surge on Developer Platforms

Chinese AI models have moved from marginal to majority usage on two major developer platforms that serve as gateways for companies accessing multiple model providers. On OpenRouter, Chinese models accounted for 57% to 67% of tokens consumed during the week of September 14, according to usage data shared with CNBC. That represents a sharp increase from just 6% to 13% in February 2026.

A similar trajectory appeared on Vercel, where Chinese models reached 55% of token usage in August, up from 11% in January. OpenRouter's dataset encompasses businesses across the United States, Europe, and a grouping of 82 nations spanning Central and South America, Africa, and Asia that the platform calls the Global South. Vercel did not disclose geographic details for its figures.

Coding and Agentic Capabilities Drive Adoption

Peter Walker, head of insights at OpenRouter, told CNBC that Chinese open-source models released this year have reached a level where they reliably execute complex agentic workflows, particularly software development tasks — a capability absent as recently as late 2025. He added that these models offer substantially lower costs than most offerings from American laboratories.

The performance gains align with recent releases from Chinese companies including DeepSeek, Z.ai and Alibaba, which have published new models with significant improvements in coding and other tasks. While the most advanced U.S. models still lead most benchmarks and attract more overall spending, the gap has narrowed sufficiently for Chinese models to capture volume-based usage on developer platforms.

Congressional Scrutiny and Geopolitical Concerns

The adoption surge is drawing attention in Washington, where two U.S. House committees are investigating the impact of rising Chinese AI model usage. Lawmakers have expressed concern about data security, technology competition, and the potential for Beijing to gain geopolitical leverage through AI integration.

Daniel Remler, a senior fellow in CNAS's technology and national security program, told CNBC that Chinese AI poses genuine economic and security threats to the United States. He warned that embedding Chinese models into foreign technology stacks risks creating durable geopolitical alignments anchored in Beijing's infrastructure.

The U.S. has sought to preserve its AI lead through export controls restricting Chinese companies' access to advanced chips, though concerns persist about remote access via overseas data centers and model distillation techniques.