Chinese artificial intelligence models are rapidly gaining acceptance among U.S. businesses as companies seek to reduce soaring AI costs without sacrificing performance, marking a significant shift in a market long dominated by American technology firms.
Developers and businesses are increasingly turning to open-source and open-weight AI models from Chinese companies such as DeepSeek, Z.ai and Alibaba’s Qwen, attracted by systems that many say now deliver capabilities approaching those of leading U.S. models at a fraction of the cost.
The trend is emerging at a sensitive moment for the United States, as the Trump administration weighs tighter oversight of advanced AI technologies while also grappling with the growing global influence of Chinese AI developers.
Industry data suggests the shift is no longer confined to experimentation. According to OpenRouter, a platform that allows developers to access and compare AI models from multiple providers, more than 30% of tokens used by U.S. companies each week since February 8 have been processed through Chinese AI models. At one point, that share climbed to 46%.
The figures represent a dramatic change from previous usage patterns.
Over the preceding 12 months, Chinese models accounted for an average of just 11% of OpenRouter’s token usage, while their share fell to only 4.5% during the first half of 2025.
The sharp increase shows how quickly developers are reconsidering the economics of artificial intelligence as operating costs become a larger concern. Early enterprise AI adoption was largely driven by access to the most capable models available, regardless of price. Increasingly, companies are evaluating whether premium AI systems justify their significantly higher operating costs.
Kyle Chan, a fellow at the John L. Thornton China Center at the Brookings Institution, said rising prices at American AI companies are changing purchasing decisions.
“Chinese AI models are particularly attractive to American companies now as AI costs skyrocket,” Chan told CNBC.
“Where previously U.S. companies were prioritizing AI adoption regardless of model, now they’re getting more cost-conscious.”
That shift is disrupting the status quo.
Many of the newest Chinese AI systems are distributed as open-source or open-weight models, allowing developers to inspect, customize, or build applications using technology that is not fully locked behind proprietary platforms. This contrasts with many flagship models from OpenAI, Anthropic and Google, whose internal architectures, training methods and core technologies remain proprietary.
The flexibility of open models has become attractive for businesses seeking greater control over their AI infrastructure while reducing dependence on commercial application programming interfaces (APIs).
The cost savings can be substantial.
According to Justin Summerville, who works on data and analytics at OpenRouter, leading Chinese open-source models are typically between 60% and 90% cheaper than comparable offerings from OpenAI and Anthropic.
Those economics are beginning to influence real business decisions. AI startup Lindy recently migrated all of its AI workloads from Anthropic’s Claude models to DeepSeek, one of China’s fastest-rising AI companies.
DeepSeek attracted global attention in early 2025 with a highly competitive reasoning model before introducing another major model upgrade in April.
Lindy’s Chief Executive Officer, Flo Crivello, said the transition immediately transformed the company’s operating costs.
“We did it, and you could see that cost curve go down, like, crash to the ground,” Crivello told CNBC.
He estimated the move would save the company millions of dollars within a matter of months.
The growing adoption extends beyond DeepSeek. Developer platform Vercel reported that DeepSeek significantly increased its share of AI token usage between May and June.
Even more striking has been the rapid rise of Z.ai’s GLM 5.2 model. Released in June, GLM 5.2 recorded the fastest adoption of any AI model tracked by Vercel during 2026.
According to Harpreet Arora, the company’s Head of Agentic Infrastructure, daily token volume surged approximately 27-fold during the model’s first full week after launch, while the number of customers using it increased about 80 times.
Arora said economics, rather than ideology, is increasingly determining which models companies deploy.
“Price is doing the work here,” he 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.”
This shows that companies are now routing different tasks to different models depending on complexity, accuracy requirements and cost, rather than relying on a single AI provider. Routine customer support, document processing, and software development tasks may be assigned to lower-cost models, while more demanding reasoning or research tasks continue to use premium frontier systems.
The approach allows organizations to reduce AI expenses while maintaining performance where it matters most. LaunchLemonade, an AI platform serving regulated industries, has observed the same trend.
Although Anthropic’s Claude and OpenAI’s ChatGPT remain its most widely used models, Z.ai’s GLM 5.2 has already entered the platform’s five most-used AI systems.
Chief Executive Officer Cien Solon said businesses are becoming increasingly pragmatic.
“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,” Solon told CNBC.
“Businesses with more mature AI strategies are increasingly willing to use them where they make technical or commercial sense.”
The growing interest is not driven by price alone. Researchers say Chinese AI models are closing the performance gap with the industry’s leading American systems.
Chan estimates that China’s most advanced models now trail the top U.S. frontier models by approximately six to nine months while costing only a fraction as much to operate.
“The new open-source models are performing well and prove capable for all but the most complex LLM tasks,” Summerville said.
Independent benchmarks increasingly support those assessments. On one closely watched benchmark measuring autonomous AI agent performance, GLM 5.2 finished within roughly one percentage point of Anthropic’s Opus 4.8 while operating at around one-fifth of the cost.
Some researchers have also reported that GLM 5.2 performs competitively with leading U.S. models on cybersecurity benchmarks, an area traditionally viewed as one of the most technically demanding applications of generative AI.
Lindy’s experience echoed those findings.
Crivello said migrating to DeepSeek V4 improved performance across many of the company’s core AI applications, demonstrating that lower cost did not necessarily require sacrificing capability.
The rapid rise of Chinese AI is also complicating U.S. technology policy. As Washington considers tighter controls on advanced AI systems, Chinese open-source models remain widely accessible around the world.
At the end of June, OpenAI delayed the rollout of a new family of models following requests from the U.S. government. During the same period, export restrictions affecting Anthropic’s cybersecurity-focused Mythos and Fable models were lifted after months of negotiations between the company and the Trump administration.
Those policy debates reflect broader concerns about maintaining U.S. leadership in artificial intelligence while limiting the international availability of the country’s most advanced technologies.
Yet some researchers warn that restricting American AI too aggressively could unintentionally strengthen overseas competitors.
Yacine Jernite, Head of Machine Learning at Hugging Face, said businesses increasingly want AI systems that they can modify, deploy independently and control without relying entirely on commercial providers.
“We’re seeing companies increasingly motivated to turn to cheaper AI stacks they can control and adapt themselves, and given the state of open-source and open-weight models that often means leveraging Chinese options,” Jernite told CNBC.
He cautioned that enterprises could eventually face an uncomfortable choice.
“There is a real risk that users get 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.”
That tension highlights the next phase of the global AI race. While American companies continue to lead in developing the world’s most advanced frontier models, Chinese developers are steadily narrowing the capability gap while competing aggressively on price. For businesses focused on controlling costs rather than on possessing the absolute best-performing AI, that combination is proving increasingly difficult to ignore.
Like this:
Like Loading...