Every so often, a new Chinese model arrives to reset the conversation about who leads the artificial intelligence race. DeepSeek did it first, startling commentators who had assumed that American export controls guaranteed a durable lead. Now Kimi 3 has revived the same chorus, with observers arguing that China has quietly closed, or even erased, the gap in frontier technology. The instinct to treat each headline release as proof of a broader technological ascent is understandable. It is also premature. A handful of impressive language models says little about the deeper architecture of Chinese innovation, and recent patent research, which captures invention across an entire economy rather than a few celebrated products, tells a more complicated story.
Indeed, China’s technology has improved. The volume of patents filed by Chinese entities has grown at a pace that dwarfs anything seen in the United States, and the country’s semiconductor sector, long dismissed as derivative, has registered genuine gains in manufacturing capacity and design sophistication. The point worth resisting is not that progress has occurred but that scale is being mistaken for depth, and depth is precisely what researchers find wanting.
Consider the chip industry, the sector most often invoked as evidence of China’s arrival as a technological power. A study examining patents granted by the United States Patent and Trademark Office in the panel display and thin-film-transistor category, an area the Made in China 2025 strategy has singled out for priority development at both the national and provincial level, offers a useful test case. Using a modified forward citation index, a measure that counts how often a patent is later cited by others once self-citations are stripped away, the research finds that a Chinese chip patent generates measurably fewer downstream citations than a comparable patent from a non-Chinese origin.
According to the study, a Chinese patent’s expected citation count falls by a factor of 0.573 compared with its non-Chinese counterpart, a gap that holds even after controlling for the age of the patent, the number of claims, and the number of assignees and inventors. Japanese and South Korean patents in the same technology class were cited more than twice and 1.6 times as often, respectively, than their Chinese equivalents.
Forward citations matter because they are the closest thing patent analysis has to a measure of genuine influence. A backward citation, the list of prior work an applicant cites, can be inflated or manipulated by firms eager to pad their own portfolios. A forward citation, by contrast, is largely beyond the applicant’s control. It reflects what later inventors, working independently, judged worth building upon. When Chinese patents accumulate fewer of these citations than their foreign peers, the implication is not that China’s engineers have stopped working. It is that their output has not yet become the kind of foundational knowledge that shapes what comes next elsewhere in the world. As the research concludes, the quantitative surge in Chinese semiconductor patents has not translated into a gravitational pull on foreign industry, and the technological gap with advanced countries will take longer to close than the raw patent counts suggest.
This pattern is not confined to chips. A separate study built a new measure of innovation capability, designed to correct for both the sheer quantity of patents a country files and the uneven quality hidden within that quantity, and applied it across the manufacturing economy as a whole. Using this framework across twenty-two manufacturing industries and more than three decades of USPTO data, the research found that only two of those industries—computer and electronic products manufacturing and electrical equipment manufacturing—actually narrowed their gap with global leaders.
Every other industry examined either fell further behind or remained stuck near the global average. Out of twenty-two sectors that China has poured subsidies, tax incentives, and political attention into for decades, the study identifies only two as showing real convergence. That ratio is a useful corrective to any narrative built around a single splashy product launch.
However, the most striking evidence comes from research that analyzes the full text of Chinese patents through machine learning rather than relying on citation counts alone. The researchers built a statistical model that treats each patent’s abstract and claims as evidence of where the “state of knowledge” in a technological field is heading, allowing them to estimate how much any given patent actually shifted that frontier. Their conclusion is unambiguous: Chinese patenting has become narrower and less innovative over time, and the influence of the country’s most important patents on future innovation has been steadily declining even as the raw number of filings has exploded.
The literature points to two forces driving this decline. Patenting activity has become more crowded, with the average number of competing filings in a given technology class roughly sevenfold higher in 2019 than in 2005, diluting the influence any single patent can hope to exert. At the same time, the underlying rate of genuine conceptual change within Chinese patenting has itself slowed. The same body of research also found that reliance on knowledge originating outside China—once a significant driver of domestic innovation—has fallen sharply, and that secondary technology classifications, the kind of cross-pollination between fields that often signals genuine creative synthesis, have become steadily less important to what gets invented.
Yet this is not necessarily a bleak picture. The same research notes that the absolute number of breakthrough patents, those in the highest percentiles of estimated importance, has continued to grow, even if the rate of that growth has slowed considerably since the mid-2000s. And the two manufacturing sectors identified as having closed the gap with global leaders happen to be precisely the ones most relevant to the current wave of AI enthusiasm, electronics and computing hardware. It is possible, then, that products like DeepSeek and Kimi 3 are real symptoms of concentrated strength in a narrow slice of the economy, rather than evidence of a broader transformation.
That distinction is the one missing from most commentary on China’s AI moment. A country can produce a handful of world-class models while its patent system, taken as a whole, grows wider in volume and narrower in ambition. Subsidies can inflate the number of filings without inflating the ideas behind them, a dynamic that fits naturally with a pattern of state-directed innovation policy, where, as one study notes, the incentive to file often outruns the incentive to discover. Impressive as the latest Chinese models may be, they do not, on their own, overturn what a decade of patent research has been showing. China’s innovation engine is louder than it has ever been. Yet, whether it is actually accelerating is a separate question, and the evidence so far suggests the honest answer is not as much as the headlines imply.
