Chinese AI Models Gain Traction with U.S. Companies Amid Rising Costs

Chinese-developed artificial intelligence models are rapidly gaining favor with U.S. companies, not just for their competitive performance but also for their significantly lower operational costs. This trend signals a potential shift in the global AI landscape, as these models close the gap with established American rivals.

Recent model releases from Chinese tech firms, including DeepSeek and Z.ai, are now being recognized for their robust capabilities, rivaling even the most advanced systems from U.S. giants like Anthropic and OpenAI. This surge in performance comes at a time when many U.S. AI labs are experiencing escalating token prices for their leading models, forcing businesses to confront unexpectedly high expenditures for AI integration and deployment.

Data from OpenRouter, a platform that aggregates access to a diverse array of AI models for developers, illustrates this growing adoption. The share of tokens utilized by U.S. companies on Chinese AI models has consistently exceeded 30% weekly since February 8th, peaking at an impressive 46%. This stands in stark contrast to the preceding twelve months, where the average stood at a mere 11%, and just 4.5% in the first half of 2025, underscoring a dramatic recent acceleration in uptake.

The ascent of Chinese open-source and open-weight models arrives as the U.S. administration intensifies its focus on regulating its most powerful AI technologies and explores strategies to manage the rapid influx of international alternatives. Regulatory scrutiny on AI development and deployment is becoming a central theme for policymakers worldwide.

“Chinese AI models are particularly attractive to American companies now as AI costs skyrocket,” noted Kyle Chan, a fellow at the John L. Thornton China Center at the think tank Brookings. “Where previously U.S. companies were prioritizing AI adoption regardless of model, now they’re getting more cost-conscious. The economics are simply becoming too compelling to ignore.”

As businesses increasingly leverage AI for new product development and internal operational efficiencies, engineers are actively exploring cheaper open-source and open-weight models. Among these, the most performant options are increasingly originating from China.

Open-source and open-weight models offer developers greater transparency and flexibility, allowing them to inspect, utilize, and sometimes adapt various components of an AI model. This contrasts with closed systems, such as many flagship models from OpenAI, Anthropic, and Google, where the underlying code and internal mechanisms remain proprietary and inaccessible.

In a significant move, the AI startup Lindy transitioned 100% of its traffic from Anthropic’s Claude models to DeepSeek in June. DeepSeek, a Chinese company that made a notable impact with a groundbreaking release in early 2025 and followed up with a new model in April, has become a key player.

“We made the switch, and you could see the cost curve plummet,” stated CEO Flo Crivello. He anticipates this strategic shift will yield millions of dollars in savings for Lindy within months, highlighting the direct commercial impact of choosing more cost-effective AI solutions.

DeepSeek witnessed a substantial increase in its share of gateway tokens on Vercel, a platform facilitating the deployment and operation of applications and websites, between May and June. This growth trajectory points to increasing developer confidence and usage.

Z.ai’s GLM 5.2 model, launched with considerable acclaim in June, demonstrated the fastest adoption rate among all models tracked by Vercel in 2026, according to Harpreet Arora, head of agentic infrastructure at Vercel. “In its first full week post-launch, daily token volume surged approximately 27-fold, and the number of customers utilizing it grew by about 80-fold,” Arora reported.

“Price is the primary driver here,” Arora emphasized. “When a task doesn’t demand peak performance, teams are increasingly opting for the most cost-effective ‘good enough’ solution, and the recent wave of models emerging from China is winning this critical trade-off.”

Justin Summerville, who manages data and analytics at OpenRouter, confirmed that open-source Chinese models can be “60% to 90% cheaper” than leading models from Anthropic and OpenAI, presenting a significant economic advantage for businesses looking to scale their AI initiatives.

The rise of these models is also influencing platforms catering to specialized industries. While Claude and ChatGPT continue to dominate usage on LaunchLemonade, an AI agent platform for regulated sectors, GLM 5.2 has quickly entered the top five models on the platform, as per Cien Solon, CEO and founder of LaunchLemonade. “Chinese models like Z.ai and Alibaba’s Qwen are becoming viable options for companies as they offer an attractive combination of performance and cost for specific workloads,” Solon commented. “Businesses with more mature AI strategies are increasingly willing to adopt them where they offer tangible technical or commercial benefits.”

Beyond cost savings, the performance benchmarks of Chinese AI models are also showing remarkable improvement.

“While often a fraction of the cost of U.S. rivals, they are now operating close to the top American frontier models,” observed Brookings’ Chan. He estimates that these models are currently “six to nine months” behind the leading U.S. competitors in terms of cutting-edge capabilities.

“The new open-source models are performing exceptionally well and are proving capable for all but the most computationally intensive LLM tasks,” added Summerville. This suggests that for a wide range of practical applications, the performance differential is becoming negligible.

On one widely scrutinized agentic benchmark, GLM 5.2 achieved performance within a single percentage point of Anthropic’s Opus 4.8, at approximately one-fifth of the cost. Further indicating their growing prowess, some researchers have reported that GLM 5.2 can match the performance of top U.S. labs on specific cybersecurity benchmarks, demonstrating its versatility and power.

For Lindy, switching to DeepSeek V4 resulted in improved performance across many core use cases, as Crivello highlighted. The ability to achieve superior results while simultaneously reducing costs is a powerful combination for businesses aiming for operational excellence.

“We’re seeing companies increasingly motivated to adopt more cost-effective AI stacks that they can control and adapt themselves. Given the state of open-source and open-weight models, this often means leveraging Chinese options,” stated Yacine Jernite, head of machine learning at Hugging Face. This trend underscores a growing desire for autonomy and customization in AI deployment.

“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,” Jernite warned. This highlights the complex strategic decisions businesses face as the AI market evolves.