U.S. companies are routing a growing share of AI workloads through Chinese-built models as token prices climb at American labs like OpenAI and Anthropic. The share of tokens processed on Chinese models via OpenRouter has held above 30% each week since early February 2026, peaking at 46%, compared with an 11% average across the prior 12 months.
Cost pressure reshapes model selection
Companies that once prioritized AI adoption regardless of vendor are now scrutinizing inference costs. Open-source Chinese models run 60% to 90% cheaper than leading Anthropic and OpenAI systems, according to Justin Summerville, who works on data and analytics at OpenRouter. That gap has widened as U.S. labs raise token prices for their most advanced models.
"Chinese AI models are particularly attractive to American companies now as AI costs skyrocket," Kyle Chan, a fellow at the Brookings Institution's John L. Thornton China Center, told CNBC. "Where previously U.S. companies were prioritizing AI adoption regardless of model, now they're getting more cost-conscious."
The shift comes as the U.S. administration tightens scrutiny of the most powerful domestic models. At the end of June, OpenAI said it would limit the rollout of a new set of models at the government's request. Export controls on Anthropic's Mythos and Fable models were also lifted that month after a standoff between the Trump administration and the company.
Adoption accelerates across developer platforms
In June, AI startup Lindy moved 100% of its traffic from Anthropic's Claude models to DeepSeek, the Chinese company that launched a new model in April. "We did it, and you could see that cost curve go down, like, crash to the ground," CEO Flo Crivello told CNBC. He said the decision will save Lindy millions of dollars within months.
Developer platforms are tracking the swing in real time. Z.ai's GLM 5.2, released in June, saw the fastest adoption of any model tracked by Vercel in 2026. Harpreet Arora, head of agentic infrastructure at Vercel, said daily token volume grew roughly 27x in its first full week, while the number of customers using it grew about 80x. The growing enterprise interest in these models has also driven demand for Deepseek AI Training among engineering teams evaluating open-weight alternatives.
"Price is doing the work here," Arora said. "When a task doesn't need the best model, teams are beginning to route it to the cheapest one that's good enough, and the recent wave of models coming out of China is winning that trade."
On LaunchLemonade, an AI agent platform for regulated industries, GLM 5.2 now ranks among the top five models, even as Anthropic's Claude and OpenAI's ChatGPT still lead in overall usage. "Chinese models like Z.ai and Alibaba's Qwen are becoming options for companies as they offer an attractive combination of performance and cost for specific workloads," said LaunchLemonade CEO Cien Solon.
Performance narrows the frontier gap
Chinese models are also closing the capability gap. Brookings' Chan estimates they currently sit six to nine months behind top U.S. frontier systems while operating at a fraction of the cost. GLM 5.2 landed within a percentage point of Anthropic's Opus 4.8 on one closely watched agentic benchmark, at roughly one-fifth the cost. Some researchers report the model performs on par with top U.S. labs on certain cyber benchmarks.
Summerville at OpenRouter said the new open-source models "prove capable for all but the most complex LLM tasks." That threshold covers a large share of commercial workloads where reliability at lower cost matters more than marginal accuracy gains.
"We're seeing companies increasingly motivated to turn to cheaper AI stacks they can control and adapt themselves," said Yacine Jernite, head of machine learning at Hugging Face. He warned that users face a real risk of being stuck "having to choose between performant but expensive U.S. proprietary models whose price and accessibility can quickly fluctuate, or using Chinese models as the only feasible alternative whenever they want to control costs or own their AI stack."
Why this matters for IT, research, and finance professionals
The cost differential is large enough to change build-versus-buy calculations across the stack. For engineering and IT leaders, the data from OpenRouter and Vercel confirms that Chinese open-weight models are already in production at meaningful scale inside U.S. companies. Ignoring them as an option means leaving 60% to 90% in inference savings on the table for a wide range of tasks.
Finance and operations teams should track token pricing trends the way they track cloud infrastructure spend. When a startup like Lindy can cut its AI costs to near zero by switching model providers, unit economics shift across every product that depends on API calls to a frontier lab. Research teams, meanwhile, gain access to models that sit within striking distance of the frontier at costs that make large-scale experimentation feasible without escalating budget approvals.
Your membership also unlocks:
