U.S. Businesses Turn to Chinese AI Models as Cost Pressures Mount
Rising token costs are pushing enterprises away from US models and towards open-weight models, changing how organizations build their AI stacks.
News
- U.S. Businesses Turn to Chinese AI Models as Cost Pressures Mount
- Stack Overflow's Decline Signals a Bigger AI Knowledge Problem
- AI Token Costs Must Drop 90% to Scale Enterprise Adoption: Palo Alto CEO
- Saudi Enterprises Rebuild Technology Stacks as AI Adoption Accelerates: Nutanix
- AI Dispatch: Bets and Warnings Define the Week
- Saudi Payment Disruptions Test UAE–Saudi Trade Ties
[Image: Chetan Jha/MITSMR Middle East]
Chinese AI models are no longer being evaluated simply as low-cost alternatives. They are increasingly becoming strategic options for U.S. companies seeking to reduce AI spending without sacrificing acceptable performance, highlighting a shift in how enterprises think about deploying generative AI, according to the Financial Times.
Developers and businesses are increasingly routing workloads to Chinese-built open-source and open-weight models from companies such as DeepSeek and Z.ai as the economics of large language models come under greater scrutiny. The trend suggests that enterprise AI adoption is entering a new phase where cost efficiency, rather than access to the most advanced frontier model, is becoming a defining consideration.
Data from OpenRouter shows that since early February, Chinese models have accounted for more than 30% of weekly token usage by U.S. companies on the platform, reaching peaks of 46%. By comparison, they averaged just 11% over the previous year and only 4.5% during the first half of 2025.
The growing adoption comes as many U.S. AI providers raise prices for their most capable models, increasing operational costs for organizations embedding AI into products, customer service, software development, and internal workflows.
That shift is encouraging enterprises to diversify beyond proprietary systems from OpenAI, Anthropic, and Google. Instead, developers are increasingly experimenting with open-source and open-weight models that provide greater transparency and flexibility while dramatically reducing inference costs.
The financial impact can be substantial. AI startup Lindy recently migrated all of its traffic from Anthropic’s Claude models to DeepSeek. Chief executive Flo Crivello said the move caused operating costs to “crash to the ground” and is expected to save the company millions of dollars within months.
Similar patterns are emerging across developer platforms. Z.ai’s GLM 5.2 became the fastest-adopted AI model on its platform in 2026. During its first full week after launch, daily token volume increased roughly 27-fold, while the number of customers using the model grew approximately 80%.
Justin Summerville, who works on data and analytics at OpenRouter, estimates that leading Chinese open-source models can be between 60% and 90% cheaper than comparable offerings from OpenAI and Anthropic.
Performance is also narrowing as a differentiator. Analysts estimate that China’s leading AI models now trail the most advanced U.S. frontier systems by roughly six to nine months.
The trend presents both a commercial opportunity and a strategic challenge. For enterprises, the emerging question is no longer whether the world’s best AI model is American or Chinese. Instead, it is whether paying a premium for frontier performance delivers enough business value to justify the additional cost.
