Chinese AI Gains Ground at US Companies as Low Prices Drive Adoption
2026-07-10
Use of Chinese artificial intelligence (AI) tools is spreading at US companies. Some models are available at about 1% of the cost of high-end US AI, helping adoption as they are embedded in business software and other applications. In services aimed at corporate users, the share of Chinese AI use is nearing 50%.
Usage has topped 30% since February
According to US startup OpenRouter, which allows users to switch among AI models from different companies, the share of US companies using Chinese AI has exceeded 30% on a token basis, a measure of data processing volume, since February and reached as much as 46%. That marks a sharp increase from the 4% range in the first half of 2025.
Rising costs make it an option
The backdrop is rising AI usage costs. High-performance models from OpenAI and Anthropic in the US have become more capable, enabling automation of long and complex tasks. But as companies expand use in internal operations and customer-facing services, token consumption rises and fee burdens increase.
On OpenRouter, the price per 1 million output tokens is $25 for Anthropic's Claude Opus 4.7, while China's DeepSeek V4 Flash costs $0.18, or less than 1%. At the company, which has 8 million users, mainly engineers, that price gap is prompting switches.
US regulations and uncertainty in China
US AI startup Lindy switched in June the model used in its service that automates email and meeting note creation from Anthropic to DeepSeek. Chief Executive Officer Flo Crivello said performance was unchanged and the move would cut costs by millions of dollars over several months.
Harpreet Arora of US AI development infrastructure firm Baseel said companies are increasingly inclined to choose models with sufficient performance for the task rather than the highest possible performance.
At the same time, US government restrictions are part of the shift toward China. In June, Anthropic received instructions from the US government to restrict use by foreign nationals and temporarily suspended provision of its 'Fable' model, which has performance equivalent to its advanced 'Mythos' model.
Kyle Chan, a researcher at the Brookings Institution, said US export controls have led to broader views that US-made models cannot always be used stably. He said many Chinese AI models are open models that can be copied, and if downloaded and used on in-house servers, the risk of sudden unavailability is smaller.
Still, there is no guarantee Chinese AI will remain freely usable. Reuters reported that Chinese authorities are discussing with companies developing advanced AI, including Alibaba Group and ByteDance, a plan to restrict overseas use of advanced AI models. Open models are also reported to be included. Companies are now at a stage where they must weigh cost cuts against geopolitical risk when deciding whether to adopt them.
Enjoyed this article? Share it with your network!