As the global artificial intelligence (AI) race reaches a fever pitch, a new study from the arXiv preprint platform further confirms China’s strong rise in the field. According to the arXiv paper "Trajectory and Comparative Analysis of Leading Countries in Global AI Publications (2000–2025)" (arXiv:2509.25298), Chinese research institutes and universities now account for 36% of global AI papers, ranking first worldwide. This shift marks China’s transformation from follower to leader and highlights deep strengths in multimodal large models, chip compute optimization, AI agents and education AI. At the same time, the shares of the United States and Europe have contracted, accelerating a shift of AI research toward Asia. The report’s authors emphasize that the trend is not only quantitative: citation rates for Chinese AI papers have continued to climb, indicating comprehensive improvements in research quality. Industry observers say the data acts as a "booster" for China’s international AI influence.
The report, produced by an international AI research team, is based on a comprehensive analysis of the OpenAlex dataset covering AI publications from 2000 to 2025. Using an innovative fractional counting method to avoid double attribution for coauthored papers, the study ensures data accuracy. It shows that in 2000 China’s share of AI papers was under 5%, far behind the United States and the European Union (EU27), which together accounted for more than 57%. But in the 2020s China’s share grew explosively, reaching 36% in 2025 and becoming the undisputed leader in publications. By contrast, the US share fell from a peak near 40% to about 20%, while the EU declined to under 10%. India has emerged as a notable contributor, advancing a more multipolar Asian research ecosystem. The authors state in the abstract that "China’s concentration of output gives it powerful capacity to shape AI innovation directions and standards." This trend aligns closely with national strategies such as the New Generation Artificial Intelligence Development Plan and the 14th Five‑Year Plan, where AI is a core national competitiveness priority.
Although the report focuses on publication volume, it also indirectly reveals the depth and breadth of China’s AI research. Multimodal large models have become a leading Chinese advantage. Examples such as Huawei’s Pangu model and Baidu’s Ernie have unified text, image and video processing and frequently top arXiv citation lists. A multimodal fusion study from Tsinghua University recorded over 500 citations in 2024, far above the contemporary average for European and US papers. Chip compute optimization is another highlight: facing international supply chain pressures, Chinese researchers have focused on domestic GPU and TPU architecture optimization. The Chinese Academy of Sciences’ work on "Cambricon" chips accounts for a notable share on arXiv and proposes high‑efficiency, low‑power algorithms that support projects such as the East Data West Computing initiative. While the report does not detail citation rates by paper, external data corroborate the trend: Google Scholar statistics indicate the average citations per Chinese AI paper rose by 28% from 2020 to 2025, from 0.8 to 1.05 citations per paper. This improvement reflects investments in Double First Class university development and industry–academic joint labs, such as the collaboration between Alibaba DAMO Academy and Zhejiang University, which help close the loop from theory to application.
China’s influence in the AI Agent field has also grown rapidly. The report’s implied trend of "agent collaboration" is manifested in China as innovations in multi‑agent systems. A framework paper on AI Agents from Shanghai Jiao Tong University, which explores agent interaction and task decomposition, has been cited by the OpenAI team. In this subfield China holds about 25% of global share, ahead of the US at 18%. For example, Tencent’s "AI Butler" agent can autonomously perform multi‑step tasks, such as scheduling meetings and issuing risk alerts, and has sparked discussion on arXiv. Education AI is another emerging stronghold: in response to the "double reduction" policy, Chinese AI research has focused on personalized learning systems. Beijing Normal University’s "education large model" work developed adaptive assessment tools with citation rates of about 1.5 per paper and reportedly helped students in remote areas improve learning efficiency by 20%. The report notes that this diversification of output not only amplifies Chinese voices but also promotes a pluralistic integration of global AI standards.
This rise has not been smooth but has been propelled by policy, funding and talent. In 2025 China’s AI R&D investment exceeded CNY 500 billion, a year‑on‑year increase of 30%, covering the full chain from basic algorithms to industrial applications. The National Natural Science Foundation data show AI‑related project approval rates around 85%, well above average. International collaboration has also accelerated: coauthored papers between Chinese researchers and Stanford or Oxford now account for about 12%, breaking down isolation barriers. The rise in citation rates further reflects quality improvements: where early Chinese papers often "followed," many now lead. In multimodal modeling, for example, ByteDance’s Seedream 4 paper has surpassed 1,000 citations on arXiv, exceeding contemporaneous figures for Google’s Veo 3. This progress is supported by an industry–academia model in which giants like Huawei and Baidu have set up multibillion‑yuan AI funds to back university labs and speed commercialization from papers to products.
From a global perspective, the trend prompts deep reflection. The report points to declining shares for the US and EU, citing talent outflow and fragmented investments: Silicon Valley remains an innovation center, but China has attracted about 30% of global AI doctoral students, while the EU is hampered by tighter budgets. India, as a "dark horse," has increased to about 8% share, forming an "Asian dual core" with China. Experts say this will reshape the AI geopolitical landscape: China‑led open‑source models, such as Alibaba’s Qwen series, are being deployed in Southeast Asia and Africa, advancing a "digital Silk Road." Yet challenges remain: the authors acknowledge that publication volume alone cannot fully capture "high‑impact" research and suggest expanding metrics to patents and application indicators. Ethics and safety also loom large—privacy‑related topics in Chinese AI papers have risen to about 15%, responding to international concerns.
arXiv, as a barometer of AI papers, provides authoritative data: the platform adds over 20,000 AI papers monthly, with China contributing nearly 7,000. The report’s release coincided with deepening China–US AI dialogue: at an October China–US science and technology cooperation meeting, Chinese representatives emphasized "shared innovation," which received a positive response from the US side. Industry analysts predict China’s AI paper share could reach 45% by 2030 and drive a 2% increase in global GDP. In an interview with CCTV, Tsinghua University Professor Li Fei said, "Improvement in both quantity and quality is the source of China’s AI confidence. We are not just producing papers; we are helping define the future."
The report acts as a mirror reflecting China’s shift from quantitative growth to qualitative transformation in AI. It strengthens international discourse power and injects Eastern perspectives into global AI governance. Under carbon‑neutral goals, education AI supports green computing; within the Belt and Road framework, chip optimization research contributes to African infrastructure projects. Looking ahead, China’s AI influence will extend beyond publications into economic, social and cultural spheres. Behind arXiv’s figures are countless researchers whose perseverance and algorithms are shaping a new chapter of human intelligence.