The Growing Self-Reliance of Chinese Innovation

The Growing Self-Reliance of Chinese Innovation

Contemporary U.S. science and innovation policy has increasingly centered on a singular objective: impeding China’s capacity to develop frontier technologies. Through export controls on scientific instruments, restrictions on international collaboration, and tighter academic exchanges, policymakers aim to slow the diffusion of scientific knowledge from the West to the East. These measures operate on a fundamental premise: that Chinese innovation is not self-reliant, but remains deeply dependent on science produced in the United States.

However, this assumption has never been comprehensively tested. Researchers have long debated whether China is merely a sophisticated importer of Western ideas. Or if it has begun to build its own intellectual foundations. This paper investigates that very tension. It asks whether the "leverage" held by the U.S.—the ability to stall Chinese progress by cutting off scientific flows—actually exists in the modern era.

Testing the assumption of scientific dependence

The central question driving this research is whether Chinese technological breakthroughs are fundamentally rooted in U.S. scientific literature. If a Chinese patent relies on a breakthrough discovered in a lab in California, then restricting the flow of that knowledge should, in theory, act as a brake on Chinese industry. But if those same patents are increasingly drawing from a growing pool of Chinese-authored scientific papers, then the strategic utility of isolationist policies begins to evaporate.

To move beyond anecdotal evidence, the authors sought to quantify the "scientific ancestry" of the entire Chinese patent landscape. They aimed to determine not just whether China is innovating, but specifically where the foundational ideas for those innovations are being born.

The data bottleneck in innovation studies

Until now, measuring this dependence has been notoriously difficult due to a lack of suitable data. The most comprehensive record of Chinese innovation lies in the corpus of patents filed at the China National Intellectual Property Administration (CNIPA). While one could theoretically check if these patents cite U.S. research, CNIPA citations are heavily biased toward Chinese-language papers. They also suffer from poor coverage prior to 2010.

The standard workaround involves looking only at "patent families" (groups of patent applications filed in multiple countries for the same invention). Specifically, researchers look at Chinese patents also filed with the U.S. Patent and Trademark Office (USPTO). This method is flawed. While these families provide standardized citation data, they represent only about 1% of all CNIPA patents. Consequently, previous assessments of China's reliance on U.S. science have likely missed the vast majority of the technological landscape. This leaves the true state of scientific self-reliance unmapped.

Mapping the scientific origins of 6.7 million patents

To overcome this, the researchers developed a novel method using Natural Language Processing (NLP; a field of AI that enables computers to understand human language) to link the full corpus of 6.7 million CNIPA invention patents to the global scientific literature. Instead of relying on formal citations, which are often sparse or missing, the authors translated patent titles and abstracts into English. They then identified "scientific phrases"—specific, technical terms found in both patents and scientific papers.

By matching these phrases, the team constructed an expansive web connecting 6.4 million patents to 38 million scientific papers from the OpenAlex database [Figure 1A]. This allowed them to assign a "scientific origin" to each patent based on where the underlying phrases first appeared in the literature. The researchers didn't stop at single words. They used "n-tuples" (combinations of phrases of length 2 up to $n$) to capture highly specific, detailed knowledge. For example, a patent mentioning both "layered oxide structure" and "copper redox" creates a unique 2-tuple. This points to a much more precise scientific antecedent than either term alone .

A pivot from U.S. reliance to domestic science

The results reveal a profound structural shift in the geography of innovation. The authors find that Chinese technological innovation is becoming rapidly self-reliant. In 2000, only 1% of the scientific phrases found in Chinese patents appeared in domestically-produced articles. By 2025, that share had surged to 26% [Figure 1B]. This represents a massive expansion of the domestic scientific base supporting industry. Simultaneously, the share of phrases found in U.S.-produced articles fell from 32% in 2000 to just 17% in 2025.

Crucially, this self-reliance is most pronounced in the most complex areas of knowledge. As the length of the phrase combinations increased, the gap between Chinese and U.S. scientific influence widened. For combinations of four or more phrases—representing highly specialized, granular expertise—41% appeared in Chinese-produced articles. In contrast, only 12% appeared in U.S. articles [Figure S2].

The study further distinguishes between "old" ideas (concepts introduced more than 10 years before a patent) and "new" ideas (concepts appearing within 10 years of a patent). While the U.S. still dominates the "old" scientific vocabulary used in Chinese patents, the pattern reverses for the frontier. The share of "new" phrases originated in China rose from 7% in 2000 to 46% in 2025. During this same period, the U.S. share plummeted from 21% to 12% [Figure 3B].

The diminishing returns of knowledge restrictions

These findings suggest that the strategic landscape for U.S. policymakers has changed fundamentally. The research indicates that the "leverage" provided by restricting scientific exports is losing its efficacy. If China is now the leading originator of the new scientific ideas required for its future technologies, then blocking access to existing Western knowledge may only serve to decouple the two scientific communities. It may not actually stop China's momentum.

The implications extend across almost all strategic sectors. The authors report that self-reliance is rising in nearly all "Critical Technology Areas" (CTAs; fields deemed vital for national security) designated by the U.S. government. This includes Semiconductors, AI, and Robotics .

Only in a few fields, such as Biotechnology and Quantum Technology, does the U.S. maintain a significant scientific lead. Even in those areas, the gap is closing.

Ultimately, the paper suggests that a policy predicated on an outdated model of Chinese dependence may be counterproductive. Isolationist strategies risk slowing global scientific progress. They also erode the transparency necessary for managing global security risks. A logical next step for researchers would be to investigate whether this trend of self-reliance is accelerating in specific emerging fields like synthetic biology. Such research would clarify if the "new idea" origination rates remain this volatile in other domains.

Figures from the paper