China’s Patent Boom Reflects Broad, Improving Innovation, Study Finds

China’s patenting has grown dramatically. The central question is whether that growth reflects technological progress or a swollen count of low-value filings. In their study, Josh Lerner and Amit Seru argue that the increase is real—not merely a product of quotas, imitation, or a handful of favored companies. Their conclusion rests not on patent counts alone, but on tests of what the patents contain, how they relate to other measures of research, and who is producing them.

The scale is striking. Looking at major patents by domestic inventors from 1985 through 2023, Lerner said the researchers counted roughly 14 million Chinese patents, compared with about three to four million US patents by US inventors. In the 15 years after China established its patent office, US activity was much higher. The sharp growth in Chinese patenting came later, especially in the most recent decade of the period they studied.

A chart shown during the discussion makes the broad change visible: annual counts for China rise sharply in recent years, while US counts are steadier, both for domestic patents overall and for patents in critical technologies. The comparison needs care. As the chart’s notes specify, it compares Chinese patent publications with US patent grants, not identical stages of the patent process. The researchers’ larger analysis focuses on major patents and domestic inventors, and they repeatedly caution that a Chinese patent may not mean exactly the same thing as a US patent. Still, the upward pattern appears across all domestic patents and across the 14 technology categories identified as critical by the US Department of Defense. Those categories include fields such as AI and advanced materials, which have civilian and military uses, as well as technologies with more direct defense applications.

One reason the headline numbers can mislead is that researchers often look at Chinese inventors’ patents filed in the United States. That figure has not exploded in the same way. But many Chinese inventors patent domestically, and only a small share—Seru put the share of the relevant patents at around 3 percent—are filed in the US. Filing abroad can be costly, and firms may have little reason to pursue US protection if they do not expect to use the technology commercially in that market. The researchers also pointed to concerns about enforcing patents across jurisdictions. A count of Chinese patents in the US therefore gives an incomplete picture of Chinese domestic activity.

Patents are an imperfect measure everywhere. Some inventions are protected through trade secrets; others may not be patented at all. And patent systems produce plenty of odd or trivial claims. Lerner invoked patents for using lasers to entertain cats and for swinging sideways rather than back and forth: evidence that patent totals need a quality check, not a reason to ignore patent data altogether.

The researchers’ case for using patents is that they contain a meaningful signal about innovation and, ultimately, productivity. Prior work has found that patent data are related to future productivity, and Seru pointed to research linking patents with stock-market valuations. Patenting is also relevant in fields with possible defense applications: technologies often have both civilian and military uses, and startups in the defense sector may seek patent protection as part of building businesses.

But evidence that patents are informative in the US cannot simply be transferred to China. Patent-office practices differ, and Chinese inventors tend to cite less than their US counterparts. Citations can be sensitive to institutional practice, Lerner noted, so a comparison that relies heavily on them could mistake differences in how offices and inventors cite for differences in inventive quality. The researchers therefore looked beyond the counts and conventional citation measures. The growing availability of patent text, and of methods for analyzing it, made it possible to compare the substance of patents more directly.

The first check asks whether China’s patent totals are inflated simply because it has many more people entering science and engineering. Measured against the number of STEM graduates, Lerner said, patenting in China and the United States looks similar. That does not establish that each patent is equally valuable, but it suggests that the sheer volume is not just a matter of Chinese graduates filing patents at an unusually high rate. The scale of China’s STEM graduate population helps explain the absolute count; the per-graduate comparison addresses a different question, about how much patenting occurs relative to that pool.

The researchers also compared patenting with scientific publication. They mapped the Defense Department’s critical-technology categories to patents and to scientific papers, then examined highly cited work in leading journals. The test was whether areas with strong Chinese scientific publication also showed strong Chinese patenting. The researchers found a strong relationship across technologies and over time, similar to the relationship they see in the United States. Their claim is an association, not proof that publication causes patenting. But scientific work provides a separate indication of where activity is concentrated. Seru gave advanced materials as an example: researchers could compare Chinese papers appearing in prominent journals with Chinese patents in the same field and trace both by year.

They also used patent text to distinguish patents that are novel from those that appear influential. In broad terms, a patent looks more novel when its language differs from patents that came before it; it looks more important when later patents use language similar to it. Combining those measures gives one way to assess whether a patent is both distinctive and consequential. This is not a claim that a text measure captures every dimension of an invention’s value. It is a way to test whether the patent surge includes work that is new relative to earlier patents and that later work appears to build on.

Applying text analysis to Chinese patents required care. Seru said the researchers translated Chinese patent text into English and tested their results in different ways, including analyses that relied on Chinese and US patent text together and analyses that excluded the Chinese text. The language and translation create limitations, but he said the results were robust across these approaches. The point of these checks was not to assume that methods developed on US patent text automatically work in Chinese, but to see whether the broad results survived different ways of using the translated material.

A separate test examined the appearance of new technical phrases, or “bigrams”—pairs of words such as “machine learning”—in patents. Lerner and collaborators had previously used such phrases to study how technologies spread, including by tracking their later appearance in job announcements and earnings calls. Here, the researchers compared when novel phrases appeared in US and Chinese patents. At first, the phrases were more common in US patents; in critical technologies, Chinese patents caught up. The relative novelty of Chinese patents improved substantially in the 21st century. Lerner emphasized the appeal of this measure as a relatively direct comparison: researchers can track the share of patents using new phrases rather than relying only on a more complex model.

Taken together, these checks cut against the claim that China’s patent boom is simply a rise in derivative, low-quality filings. They answer different questions: patenting per STEM graduate puts the total in relation to the size of the research workforce; scientific publications offer a separate signal about activity by field; text measures test novelty and later influence; and bigrams track the arrival of new technical language. The researchers do not say every patent is important, or that patent data capture all innovation. Their narrower point is that they find no sign that rising volume has come with falling quality. By the measures they examined, quality has moved in the other direction.

The distribution of patenting complicates familiar descriptions of the Chinese innovation system. In critical technologies, academic institutions account for about a quarter of Chinese patents, compared with roughly 3 percent in the United States. Lerner described the Chinese university share as about eight times the US share. Universities also appear prominently among the largest patenters: seven of the top ten in China are universities, he said, while US universities rank much lower.

The technologies in this comparison are not exclusively military. The researchers matched patents to the Defense Department’s critical categories using both technology classifications and language. Those categories include dual-use areas such as AI and advanced materials, as well as fields such as storage and renewables. “Critical” in this analysis therefore does not mean that every patent has a military application.

The private sector remains a major source of Chinese patenting. Seru estimated that corporate patenting accounts for about 55 to 60 percent of activity in China, compared with roughly 90 percent in the United States. What differs is how concentrated corporate patenting is. In the US, a relatively small number of large companies account for a substantial share. In China, the activity is more dispersed. Huawei is prominent, but it does not account for the whole story: the researchers found that Chinese innovation was less concentrated among top patenters than US innovation, including when they looked at novel and impactful patents.

That finding runs against the idea that China’s progress is mainly the work of a few state-favored giants. It also complicates the view that pushing the technological frontier necessarily requires very large firms. The researchers did not claim to have established why the Chinese system produces a wider distribution. Seru said they were still considering whether the government acts as a coordinator or “matchmaker,” helping connect universities and firms, or whether incentives encourage institutions to produce patents. He presented these as hypotheses, not settled explanations. His speculation was that because resources go to state-owned enterprises while patenting is spread across universities and many firms, some coordination or incentive structure may help connect the activity. The patent data show the distribution; they do not identify the mechanism behind it.

Lerner pointed to another possible difference: the movement of people and ideas between universities and firms. Chinese academics appear to have a range of arrangements for patenting, including people who patent through a university and through a company. That kind of fluidity may make collaboration easier, he suggested. In the United States, university conflict-of-interest and compliance procedures can make such arrangements more cumbersome. Those rules exist for reasons, but they may also raise the cost of collaboration. Lerner described the contrast as suggestive rather than a complete explanation of the differences in patenting.

The role of foreign-born and returning researchers also differs from common assumptions. The researchers found that US patenting relies more heavily on immigrants than Chinese patenting does. They also found that US-trained Chinese nationals returning to China are not responsible for a large share of China’s patenting. Their contribution is significant in absolute terms, Lerner said, but small relative to the scale of China’s innovation system. Seru allowed that overseas-trained returnees may have helped build universities and other parts of the system earlier; his point was that Chinese innovation now rests overwhelmingly on homegrown talent. This finding complicates the idea that China’s progress is primarily an export of expertise trained in the United States.

For Josh Lerner and Amit Seru, the evidence changes the terms of the policy discussion. Lerner challenged the account that Chinese progress is mostly theft from US firms or the product of a few subsidized companies. Taking China’s innovation seriously, he argued, requires recognizing that the activity is broad, increasingly novel, and built substantially on domestic capabilities.

He also raised concerns about the path from university research to commercial use in the United States. As an example, Lerner described a venture fund focused on AI and health care whose founder had visited aspiring faculty entrepreneurs at Harvard Medical School. For clinical AI applications, the entrepreneurs described spending 12 to 18 months obtaining approval through the university’s institutional review process, then facing another review at the Department of Health and Human Services. In a field moving as quickly as AI, Lerner argued, a process that takes roughly two years before a first step toward market use can leave US entrepreneurs at a disadvantage. The Chinese entrepreneurs backed by the same fund faced nothing comparable, he said. The example illustrates the delay Lerner sees in commercialization: the entrepreneurs were describing time spent securing permission before they could begin moving an application toward the market.

Seru connected the administrative burden to a broader concern about universities. Universities helped build the US scientific and technological system after the earlier Sputnik moment, he said. They can have bureaucracy and activities unrelated to their core mission; that does not mean the research and collaboration infrastructure inside them should be weakened. He warned that restrictions and skepticism around university funding and collaboration can damage that capacity, which takes time to build and time to restore.

The effects of funding disruptions, Lerner said, are not easily reversed. When federal funding freezes affected Harvard labs, some laid off specialized staff. Even when funding resumed after a court ruling, the teams could not simply be reassembled. Researchers had moved to other labs, including abroad, or to the private sector. A research group, in his analogy, is not a construction site where workers can be brought back on demand. The policy concern is not only whether funding eventually returns, but whether the people and working relationships that supported research remain in place.

Immigration policy is another part of the innovation system. The researchers’ findings suggest a contrast: China’s innovation growth is largely homegrown, while a substantial share of US innovation depends on people born elsewhere. Davis summarized the US contribution from foreign talent as somewhere between 25 and 50 percent; Seru agreed that US breakthroughs often depend on joint production between US-born and immigrant researchers. The exact contribution varies, but the policy implication in their discussion was clear: restricting high-skilled immigration risks weakening a source of US strength.

Lerner described a global competition for scientific and managerial talent. At a G20 innovation ministerial he attended, countries were discussing how to attract people from around the world, including governments without the United States’ natural advantages. Against that backdrop, policies that make the US less welcoming could matter more, not less. The researchers’ comparison also makes clear that the two countries’ systems draw on different sources: China’s current patenting activity is overwhelmingly domestic, while US innovation depends more heavily on immigrants.

The speakers did not argue that every feature of the US system should be preserved unchanged. They acknowledged that universities and other parts of the innovation ecosystem need improvement. Their concern was that broad cuts, disruptions, or barriers could erode the system’s productive core at a moment when China is already close to the United States in many critical technologies and ahead in some. Seru’s closing distinction was between scrutinizing institutions for ways to improve them and undertaking changes that disrupt the system’s capacity while that competition is intensifying.