Chinese AI Models Dominate US Corporate Token Usage on OpenRouter, Complicating Restrictions

Chinese AI models now account for roughly 60% of all tokens consumed by US companies on OpenRouter, a platform that provides access to a wide range of AI models from various developers, according to a report by Nectar Gan at Bloomberg. The figure underscores the significant reliance of the US startup and tech ecosystem on Chinese made models, particularly those from developers such as DeepSeek and Z.ai. This dominance poses a dilemma for policymakers in Washington. On one hand, national security concerns and technological competition drive calls for restrictions. On the other, imposing drastic limits risks disrupting the operations of thousands of US companies that have come to depend on these models for cost efficiency and performance.

Data from OpenRouter shows that the market share of Chinese AI models surged dramatically from just about 1% in 2024 to more than 60% by mid 2026. This spike is driven by price and performance factors. Models like DeepSeek and Zhipu AI are considered capable of matching the capabilities of leading US frontier models such as GPT 4o or Claude 3.5, but at a significantly lower cost. A CNBC report from July 7, 2026, noted that the latest model releases from Chinese companies, including DeepSeek and Z.ai, are viewed as highly competitive compared with leading US frontier systems. This comes amid rising costs for using models from OpenAI and Anthropic, prompting many US companies, especially startups, to switch to cheaper alternatives.

OpenRouter itself functions as a kind of marketplace or API gateway that lets developers compare and choose AI models from various providers. The platform is popular among startups for its flexibility. With Chinese models offering highly competitive per token pricing, many US developers have organically shifted to these models to maintain their business margins. This phenomenon carries complex geopolitical implications. Washington has for several years sought to restrict China's access to advanced US made chip technology, on the assumption that this would slow China's AI progress. However, OpenRouter data instead shows that Chinese AI models have not only survived but have become the top choice in the US market itself.

Policy analysts cited in the Bloomberg report highlight that this situation places US regulators in a difficult position. If restrictions are imposed aggressively, US companies already integrated with Chinese models would be directly affected. Disruptions to AI services could impact the productivity and competitiveness of these businesses. On the other hand, allowing the situation to continue without oversight is also seen as risky from a data security standpoint and in terms of potential foreign influence. The US government now faces the question of how to balance national security interests with the needs of a domestic tech industry hungry for affordable AI solutions. Chinese AI companies such as DeepSeek and Zhipu AI continue to improve their models.

Some of their latest models have even been reported to outperform US models on certain benchmarks, while still offering lower prices. This pressures US AI companies to keep innovating and cutting costs. From an industry perspective, this price competition has triggered a price war in the global AI API market. US startups that previously relied on OpenAI or Anthropic now have more options. However, the growing dependence on Chinese AI infrastructure also raises new concerns about technology supply chains and potential vulnerabilities. The OpenRouter report itself does not provide details on the types of tokens or specific applications using Chinese models. However, in general, tokens are the basic units that AI models use to process text.

The more tokens processed, the greater the workload and cost. The 60% dominance indicates that the volume of data processing by US companies through Chinese models is very large. As of now, there has been no official statement from the White House or US regulatory bodies regarding specific steps to be taken in response to this data. Discussions among policymakers are ongoing, focusing on how to design rules that protect national interests without harming domestic innovation and business. This development is also drawing attention from the global tech industry. If this trend continues, Chinese AI models could not only dominate the US market but also potentially become the de facto standard in many other countries seeking affordable AI solutions.

This could fundamentally reshape the global AI competitive landscape.