Chinese AI Models Gain Majority Token Share on Vercel as House Committees Investigate

New Vercel figures extend a shift already visible on OpenRouter. Lower prices are winning work, though U.S. models still draw more spending.

By 3 min read
Chinese AI Models Gain Majority Token Share on Vercel as House Committees Investigate
Chinese AI Models Gain Majority Token Share on Vercel as House Committees Investigate

Listen to this story

The audio brief

About 1:36
0:001:36
Read transcript
Chinese models accounted for 55 percent of the tokens sent through Vercel in August, up from 11 percent in January. That means they now handle a majority of the work routed through a second developer platform, extending a shift already seen on OpenRouter. There, Chinese models’ share rose from 6 to 13 percent in February to between 57 and 67 percent in the week of September 14. These figures count tokens—the pieces of text models process—on platforms where developers can choose among providers. They show which models get routed work, not Chinese models’ share of all AI use. Vercel hasn’t provided a geographic breakdown. Price helps explain the change. OpenRouter’s Peter Walker said Chinese open-source models have become credible for advanced, multi-step tasks, especially coding. That’s his assessment, not proof they perform equally well on every task. Harpreet Arora of Vercel said a lower-cost model becomes compelling when it’s capable enough for the job. Companies still use leading U.S. models for some harder tasks, and those models attract more spending overall. On OpenRouter, Chinese models handled 67 percent of tokens for companies in its “Global South” grouping. U.S. companies account for about half of the platform’s tokens. Two House committees are investigating the adoption. The scrutiny comes amid U.S. chip controls and concerns about overseas chip access and model distillation—not findings about the models in these platform figures. The key question is whether rising token use brings Chinese providers more spending, too.

Story brief

3 key points

Chinese models’ share of Vercel traffic climbed from 11% in January to 55% in August, while OpenRouter recorded 57%–67% in the week of September 14. The shift suggests price and improving coding capability are redirecting developer workloads, even as companies retain U.S. models for harder tasks and spend more on them overall. Two House committees are examining the adoption’s implications. These platform token...

  1. 01

    OpenRouter’s Chinese-model share rose from 6%–13% in February to 57%–67% in the week of September 14.

  2. 02

    On OpenRouter, Chinese models handled 67% of tokens for companies in its “Global South” grouping; U.S. companies account for about half of platform tokens.

  3. 03

    OpenRouter’s Peter Walker said Chinese open-source models have become credible for advanced multi-step work, especially coding; this is an assessment, not a task-by-task benchmark.

Chinese AI models now handle most tokens used on two developer platforms, according to company-usage figures shared with CNBC. Vercel’s newly reported numbers extend a shift already visible on OpenRouter. As lower prices help win work, two U.S. House committees are investigating the impact of rising adoption.

The shift reaches a second platform

In February, Chinese models accounted for 6%–13% of tokens used on OpenRouter. Their share reached 57%–67% in the week of September 14. Vercel’s data shows that the rise is not confined to OpenRouter: Chinese models also reached a majority of tokens used on its system by August.

Both figures count tokens, the units of text models process, on platforms that give developers access to different providers. They show where work is being sent through those services, not Chinese models’ share of all AI use. Vercel did not specify a geographic breakdown, so its figures cannot show where that platform’s growth is concentrated.

Why cheaper models are getting the work

OpenRouter insights head Peter Walker told CNBC that Chinese open-source models released this year have become credible for advanced multi-step work, especially coding, in a way they were not in late 2025. He also described them as much cheaper than most models from American labs. That is his assessment, not a finding that the models perform equally well on every task.

Vercel’s Harpreet Arora described the buying decision to CNBC: once a lower-cost model is capable enough for a job, its price becomes compelling. Companies still turn to leading U.S. models for some more complicated tasks, he said. Those U.S. models also attract more overall spending, despite the token shares on these two platforms.

Where adoption is strongest

OpenRouter’s figures cover companies in the U.S., Europe and 82 countries it groups as the “Global South.” Among companies in that last group, Chinese models handled 67% of tokens in recent weeks. U.S. companies account for about half of all tokens on OpenRouter, giving the platform’s geographic breakdown a wider context than the regional figure alone.

A commercial shift draws political scrutiny

Two U.S. House committees are investigating the impact of rising Chinese-model adoption. Washington has already restricted Chinese AI companies’ access to advanced chips through export controls. CNBC also describes U.S. concerns about remote access to chips through overseas data centers and distillation, a method in which a newer model learns to mimic an established one. Those concerns are not findings about the models counted in the platform figures.

Daniel Remler of the Center for a New American Security warned CNBC that countries using Chinese models could move closer to China’s technology sphere and eventually its geopolitical orbit. That remains a forecast. The next test is whether this growing use brings Chinese providers more spending—and whether lawmakers treat adoption itself as a security concern.

Sources

  1. cnbc.comChinese AI models surge in global popularity — and Washington is worried

Loading discussion...

YOUR READING SPACE

Notifications