American technology companies are rapidly shifting AI workloads from premium US providers to cheaper Chinese alternatives, with some reporting savings of millions of dollars annually. The movement reflects a fundamental shift in how businesses evaluate artificial intelligence, prioritizing cost efficiency alongside performance as AI expenses become a major line item.

What's Driving This Sudden Shift Away from US AI Models?

The economics are stark. Chinese open-weight models, which are AI systems released publicly for anyone to download and run, cost 60% to 90% less than leading offerings from OpenAI and Anthropic, according to OpenRouter, a platform that routes AI requests to different providers. To put this in perspective, a standardized workload costs $4,811 on Anthropic's Claude, $3,357 on OpenAI's ChatGPT, but only $1,071 for DeepSeek, $948 for Moonshot's Kimi, and $544 for Zhipu's GLM. For developers, the difference is even more dramatic: an hour-long coding session that costs about $10 on Claude costs less than 50 cents on DeepSeek.

San Francisco startup Lindy.ai, which builds AI assistants for email and calendar management, offers a concrete example of this trend. The company's chief executive explained the decision bluntly.

"By far, our No. 1 expense was Anthropic," said Flo Crivello, CEO of Lindy.ai.

Flo Crivello, Chief Executive Officer at Lindy.ai

Crivello noted that DeepSeek-V4 was "just 10x cheaper" and that moving 100% of the company's traffic to the Chinese model saved Lindy millions of dollars. "So it was a very, very simple business decision," he explained. For a company with more than two dozen staff members, Anthropic had become the largest expense, ahead of payroll and rent.

How Are Major Companies Adopting Chinese AI Models?

The adoption is spreading beyond small startups. Major enterprises including Airbnb and Siemens are experimenting with moving daily operations toward Chinese AI companies like Alibaba and DeepSeek. A UBS report highlighted an unnamed major global bank that began hosting Alibaba's Qwen models to control costs while maintaining access to premium systems like Claude for tasks requiring maximum capability. Coinbase's chief executive Brian Armstrong stated that his company is experimenting with making open-weight options the default, including GLM 5.2 and Moonshot's Kimi 2.7, while allowing engineers to select models based on specific task requirements.

The adoption data reveals this is far more than isolated cost-cutting experiments. OpenRouter reported that Chinese models accounted for more than 30% of tokens (the basic units of text that AI systems process) used by US companies on its platform every week since February 8, reaching as high as 46%. This compares with a 12-month average of just 11% and only 4.5% in the first half of 2025. On Vercel's AI Gateway, a platform used by developers, daily token volume for Zhipu's GLM-5.2 rose 50-fold from mid-June, and DeepSeek V4 Flash became the platform's largest individual model by volume, taking more than 20% of traffic.

• Market Share Growth: Chinese models grew from 4.5% of token usage in early 2025 to over 30% by mid-2026, with peaks reaching 46% on some platforms
• Enterprise Adoption: Major companies including Airbnb, Siemens, Coinbase, and an unnamed global bank are integrating Chinese models into daily operations
• Ecosystem Development: Alibaba's Qwen has built the largest model ecosystem on Hugging Face with more than 113,000 derivative models, and Chinese open-source models account for 41% of downloads on the platform

What Are the Latest Chinese AI Models Offering?

Beijing-based Moonshot AI launched Kimi K3 on July 16, marking the latest high-profile release in the competitive Chinese AI landscape. The model features 2.8 trillion parameters (the adjustable weights that allow AI systems to process information) and uses a mixture-of-experts architecture, which allows different parts of the model to specialize in different types of tasks. Kimi K3 includes native visual understanding and can process a one-million-token context window, meaning it can analyze roughly 100,000 words at once. Moonshot announced that full model weights would be released by July 27, enabling users to self-host the model or run it through open-model providers.

On independent testing, Kimi K3 ranked fourth among frontier models, behind Claude Fable 5 and GPT-5.6 Sol but ahead of Claude Opus 4.8. It achieved the top score on the Arena.ai WebDev leaderboard with 1,679 points, above Claude Fable 5's 1,631 and GPT-5.6 Sol's 1,618. However, Kimi K3 is not a bargain-basement option: at $3 per million input tokens and $15 per million output tokens, it ranks around Claude Sonnet pricing, making it Moonshot's most expensive release to date.

How to Evaluate Chinese AI Models for Your Organization

• Performance Assessment: Compare benchmark scores on standardized tests relevant to your use case, such as coding tasks, knowledge questions, or visual understanding, rather than relying solely on price
• Cost-Benefit Analysis: Calculate your actual token spending and project savings by testing models on representative workloads before full migration
• Data and Hosting Considerations: Evaluate data privacy policies, hosting arrangements, and compliance requirements, particularly for sensitive information or regulated industries
• Reliability and Support: Assess uptime guarantees, customer support availability, and model update frequency to ensure consistency with your operational needs

The broader implications of this shift extend beyond cost savings. For ordinary users, model choice is likely to appear indirectly through their apps and workplace software. If application makers and workplace software firms can reduce their AI running costs, more features may become viable without steep subscription increases or strict usage limits. However, businesses will need to judge performance, reliability, hosting arrangements, and data handling carefully rather than choosing a model based on price alone.

US companies still lead in developing the most capable frontier models, with experts noting that Chinese systems remain six to 12 months behind on capabilities. However, China's strategic push into open-weight software, partly shaped by US restrictions on advanced Nvidia hardware, has created a powerful alternative for firms that do not need the absolute best model for every task. The next critical test will be whether premium American AI providers respond with lower prices, clearer value propositions, or still greater capability. For the industry, cost is increasingly becoming as important as headline benchmark performance in determining which AI systems businesses actually adopt.