China has intensified its push to challenge United States leadership in Artificial Intelligence, a goal first set in 2017 with a plan to lead the field by 2030. This year’s surge includes DeepSeek, a startup whose large language model competes with Western systems like ChatGPT and Grok while using a fraction of the computing power. Alibaba introduced a powerful new model and committed to building more global data centers, and Tencent released Hunyuan-A13B with an emphasis on speed, capability and openness for developers. Together, these moves underscore a coordinated effort among startups, tech giants and the state to narrow the gap. Nvidia chief executive Jensen Huang recently cautioned that the United States is not far ahead, noting China’s rapid adoption of new technology.
China’s advantages include a massive online population that serves as a real-time testbed and a design philosophy that prioritizes models optimized for cheaper hardware, lowering deployment costs. University of Washington professor Pedro Domingos told DW that China is now moving at the same speed as the United States and has been closer to the frontier than many assumed, pointing to early work such as Baidu’s deep learning group in 2010. Chinese firms are also leaning into open source: models like DeepSeek, Qwen-3 and Kimi K2 are released openly, in contrast to Western rivals that increasingly lock methods and capabilities behind paywalls. By July, Chinese firms had released more than 1,500 large language models, or about 40% of the 3,755 models launched globally, according to state media. On October 18, 2025, the OpenCompass leaderboard showed China holding 14 of the top 20 model spots across reasoning, knowledge, math and coding tasks, with nine Chinese entries open source versus none from Silicon Valley.
Chip restrictions remain a major headwind. United States export controls have choked access to cutting-edge semiconductors and tools, forcing a shift to older, less efficient hardware. Beijing has countered by banning some imports from American suppliers such as Micron Technology and accelerating domestic chip efforts. The constraints are spurring local innovation: Cambricon Technologies reported a 14-fold jump in quarterly revenue, after a 44-fold rise in the first half, as startups seek local alternatives. Domingos argued the export curbs are counterproductive because they incentivize breakthroughs that make advanced training possible on legacy chips.
While the United States continues to lead at the frontier of Artificial Intelligence research, including safer instruction-following systems and multimodal capabilities, China is expanding real-world impact by exporting cloud and data center infrastructure across Asia, Africa and Europe via companies like Alibaba and Huawei. Beijing is also advancing its own governance frameworks, aiming to shape global standards and promote models trained on Chinese data and values in a tightly controlled information environment. Domingos warned that whoever controls large language models shapes how history and reality are interpreted, while Robin Feldman of UC Law described the Artificial Intelligence race as a new form of Cold War, likely to be won by the country that sustains the greatest lead.
