WASHINGTON — U.S. President Donald Trump has dismissed calls to slow artificial intelligence development, citing competition with China, even as Chinese open-source AI models have come to account for more than half of global usage.
Speaking to reporters in Ireland on the 13th, Trump said the United States is ahead of China in AI and that he wants to keep it that way. He said safeguards could be put in place, but complained that "there are too many negative forces highlighting what shouldn't be highlighted" and that "they are highlighting things that will not happen."
Earlier, Anthropic Chief Executive Dario Amodei argued in a lengthy post on the 12th for slowing the pace of AI development. Key industry figures including OpenAI's Sam Altman, Google DeepMind's Demis Hassabis and Elon Musk of xAI expressed sympathy and support for moderating development speed and introducing government regulation.
Sen. John Cornyn, a Texas Republican, said in a post that AI developers are free to slow down their own product development if they wish, but that one thing is certain — competitors, including hostile states, will not do the same. Texas is the state with the most active data center industry, the backbone of AI.
Democrats, however, say they will tighten AI regulation if they win control of the House in the midterm elections. According to The New York Times, former President Barack Obama told a closed-door Democratic fundraiser in Manhattan on the 10th that AI could become dangerous if not properly controlled, and strongly urged House Minority Leader Hakeem Jeffries to establish a framework for public debate should he become speaker. Politico, a U.S. online outlet, recently reported that Democrats are looking to form a special committee on AI to oversee the technology industry if they retake the House.
Against this backdrop, global users' use of Chinese AI models has overtaken that of U.S. models. The shift reflects growing use of Chinese-built AI models to power the latest American coding agents.
Japan's Nihon Keizai Shimbun, citing U.S.-based OpenRouter, reported that Chinese AI accounted for 55% of usage in June on a token basis, compared with 43% for U.S. AI. That marks a sharp increase from September last year, when U.S. models accounted for 76% and Chinese models 20%.
Among all AI models, the most-used large language model on OpenRouter as of the 7th of this month was Chinese firm Tencent's Hy4 Preview. Released on the 28th of last month, Hy4 Preview is the cheapest of China's new flagship models, whose main strength is low pricing.
Performance has also improved from earlier versions. According to the South China Morning Post, Hy4 Preview ranked eighth in the world in web development on Code Arena, which evaluates LLM performance in real time. U.S.-based Anthropic's Claude Fable 5 ranked seventh, a comparable level.
Notably, OpenRouter data show the app sending the most traffic to the Hy4 model — 815 billion tokens in total — was Claude Code, Anthropic's own coding agent. Users of the Anthropic coding agent are not using Anthropic's Claude models for actual inference work, instead turning to the more cost-effective Hy4 model. In effect, a combination in which Hy4 serves as the brain and Claude Code as the hands and feet is spreading quickly across AI development. Chinese AI firms are also rapidly securing funding, with Zhipu AI, which operates the Kimi3 model, completing a $5 billion financing round.
