China’s role in global artificial intelligence development is shifting from follower to contender. According to the 2025 edition of Stanford University’s AI Index, Chinese developers produced 15 of the most prominent AI models in 2024, trailing only the United States’ 40. While the U.S. maintains its lead, the report reveals China’s position is no longer marginal. It now dominates in patents, research citations, and public optimism around AI.
The reportnotes that China accounted for 69.7% of global AI patent filings in 2023 and continues to lead in the number of top-cited scientific papers. Meanwhile, public sentiment remains starkly divided: 83% of respondents in China view AI positively, compared to just 39% in the United States.
Another finding points to a transformation in who builds modern AI. Just six years ago, academia contributed over half of the world’s notable models. That figure dropped to 10% in 2024, with corporate labs like OpenAI, Meta, and Google DeepMind now producing 90% of the most impactful systems. Chinese companies—including Baidu, Huawei, and DeepSeek—are rapidly filling the gaps left by U.S. dominance.
The benchmark gap is also narrowing. Meta’s Llama 3.1 405B model trails OpenAI’s GPT-4o by only 1.5 points on the MMLU benchmark, according to the 2025 AI Index. This parity reflects one of the report’s key observations: open-weight models are catching up to proprietary ones in terms of reasoning and general knowledge tasks.
With the recent release of new Llama 4 models, which were not part of the analysis, the gap between closed and open model appears to be already closing. Llama 4 Maverick, which is positioned by Meta as a high-performance multimodal model, shows great performance on visual reasoning tasks, achieving 90.0 on ChartQA and 94.4 on DocVQA, outperforming both GPT-4o and Gemini 2.0 Flash.
According to the report, organizational adoption continues to grow across regions. Globally, 78% of organizations reported using AI in 2024, up from 55% the year before. The Asia-Pacific region led the expansion, particularly in industrial robotics, where China installed over 300,000 units in a single year.
Tensions Rise Over DeepSeek
While China’s AI progress is accelerating, its growing influence is now a target of expanding U.S. scrutiny. At the center of the latest political friction is DeepSeek AI, a company based in Hangzhou whose models have gained traction as alternatives to ChatGPT, Gemini, Claude, and other leading AI models.
In January, the U.S. Navy banned the use of DeepSeek AI on government devices citing security concerns. That was followed by a ban by the state of Texas on February 3. The Trump administration is considering a broader federal ban on DeepSeek.
Facing increasing regulatory pressure, DeepSeek accelerated the launch of its R2 model in late February, a release originally planned for later in the year. The move was interpreted as a response to potential platform removals and trade restrictions.
OpenAI formally submitted concerns to U.S. regulators, warning that DeepSeek is “state-subsidized” and “state-controlled” and could be “compelled by the CCP to manipulate its models to cause harm.”
The company has also drawn criticism for its recruitment practices. According to a February report in the New York Post, DeepSeek hired multiple engineers previously affiliated with Microsoft’s now-closed Shanghai lab, raising concerns about knowledge transfer and corporate oversight.
The facility had supported 258 AI projects and trained nearly 10,000 professionals. The closure followed months of speculation after the removal of lab signage and equipment in early 2025.
It wasn’t the company’s only move. In May 2024, Microsoft asked hundreds of China-based engineers to consider relocating abroad. That summer, the company required employees in China to use iPhones due to concerns about Android security updates and device compliance.
Microsoft’s China-based joint venture Wicresoft — plans to wind down operations and lay off staff, signaling a broader corporate retreat from the region.
U.S. Export Controls Tighten on AI Chips
Meanwhile, the U.S. Commerce Department expanded AI chip export restrictions to China, targeting Nvidia’s remaining exportable chips—A800 and H800—as well as dozens of Chinese firms. The restrictions extended to equipment servicing, placing pressure on companies like ASML to stop maintaining older DUV lithography systems still in operation inside China.
This followed earlier moves to block Nvidia’s A100 and H100 chips in 2023. With EUV systems already banned from being sold to Chinese firms, the latest measures seek to restrict even legacy tools. These decisions are aimed at denying China access to the computational infrastructure needed to train large-scale models.
However, China is responding with its own playbook. Huawei developed its 7nm Kirin 9000S processor without access to EUV technology, relying instead on older, less efficient DUV processes. Although performance lags behind cutting-edge chips, the workaround highlights China’s persistence in maintaining a domestic supply chain.
Beijing also introduced export restrictions on gallium and germanium, two elements essential to semiconductor manufacturing. And a $47.5 billion national fund—referenced in both the AI Index and trade reporting—is being deployed to further expand China’s chipmaking capacity over the coming years.
Despite Washington’s aggressive export controls, some U.S. tech executives are skeptical about long-term effectiveness. “AI will continue to spread globally. No single country, including the United States, can stop this. The question is not whether AI will spread, but whether the United States and its closest allies will remain at the forefront of its development,” warned Microsoft President Brad Smith in February.
OpenAI Pushes for Federal Strategy
OpenAI has taken a proactive approach in lobbying the U.S. government. On January 13, the company released its policy blueprint, AI in America, arguing for the creation of AI Economic Zones across the country. These hubs would attract investment into critical infrastructure such as data centers, chip fabrication plants, and energy facilities.
Framing China as the primary competitor, OpenAI’s Chris Lehane warned of strategic drift. “That would just create real dissonance, both on a national security and economic competitiveness front.” The blueprint also identified nations like Saudi Arabia and the UAE as “swing states” in the AI contest—jurisdictions whose investment flows could lean either toward U.S. or Chinese platforms.
The company also backed tighter export rules on frontier AI models and technologies, while defending its own model training methods. It argues that large-scale web scraping falls under fair use in the U.S., but raises concerns about foreign firms exploiting American content without equivalent protections.
A Multipolar Future With Unresolved Risks
The AI Index shows that the AI race is becoming more globally distributed. France is launching a €109 billion AI investment initiative. Other countries like India, Brazil, and Saudi Arabia are seeing rapid growth in AI workforce development, according to Stanford’s report.
However, most frontier models still come from the U.S. and China, and transparency in responsible AI practices remains limited. Only 31% of models released in 2024 included standardized safety benchmarks, and just 18% of large companies reported conducting fairness testing prior to deployment.
As sanctions deepen, and companies reorganize around national boundaries, the global AI ecosystem is fracturing. Whether this fragmentation leads to competing standards or opens new space for cooperation is still unresolved—but one thing is clear: China’s position in the AI race is no longer peripheral.
