The World Intellectual Property Organization (WIPO) recently reported a significant increase in patent filings related to generative artificial intelligence (GenAI). To discuss the findings and their implications for innovators, patent practitioners, and the evolving AI landscape, Westlaw Today spoke with Finnegan partners and co-leaders of the AI Practice Frank DeCosta and Karthik Kumar, who shared their perspectives on the trends highlighted in WIPO’s research.
Westlaw Today: First of all, how did WIPO's "Patent Trends Update in GenAI" report come about and why are the findings significant?
Frank: This was an update to a report published by WIPO in 2024 on the generative AI landscape. Significant findings of the report include the reported growth statistics of GenAI patent publications: about 14,000 in 2023, 19,000 in 2024 and almost doubling to about 37,000 in 2025. While the rate of growth is impressive, it is important to keep in mind that this statistic only tracks published patent applications. These applications will be subject to examination by the national patent authorities and not all published patent applications will result in a granted patent.
WIPO also reports that GenAI now accounts for 8.7% of all AI published patent families, up from 6.1% in 2023. An important takeaway from this statistic is a reminder that as important as GenAI is becoming in the IP space, more than 90% of the published patent applications in the AI space are not related to GenAI. So, as we work with our clients to develop strategies for AI IP protection, we have to keep the big picture in mind to provide guidance that addresses the growing universality of AI innovation in its many forms that impact business.
WT: According to the report, what new technologies are currently dominating patent applications? When it comes to AI, has there been a shift in which technology architectures are becoming more popular?
Frank: The WIPO report credits large language models with overtaking other technologies. As reported, LLMs are better suited for text than images. It is well known in the AI tech world that whoever has access to the most data has the advantage in the sense of the ability to create more robust and accurate AI systems. The popularity of LLMs is driving a near insatiable demand for text training data. This demand has consequences that we see playing out in several areas, including, for example, privacy regulations and copyrights.
WT: Which companies are leading the charge among U.S.-based patent applicants?
Karthik: The report identifies Alphabet as the largest U.S.-based owner of GenAI patent families, followed by Microsoft and IBM. That is itself a shift from the 2024 report, in which IBM led the U.S. field. One should not read too much into the precise ordering, though, because these are cumulative counts and the recent filing rates tell a more interesting story. IBM, for example, sits high on the cumulative list but has a comparatively modest recent filing rate, while other U.S. companies are filing aggressively right now.
The entrant we find most notable is Nvidia, which appears in the top ranks for the first time. Nvidia's arrival reflects its expansion beyond GPU [or graphics processing unit] hardware into AI software, frameworks and model architectures — in other words, a company known for the "picks and shovels" of the GenAI boom is now building a portfolio around the models themselves. Adobe has likewise continued to build out its GenAI portfolio, consistent with its focus on content-generation tools. More broadly, the U.S. figures reflect a steep recent acceleration — faster than China's rate over the same period — even though China remains far larger in absolute terms.
WT: How are companies outside of the U.S. progressing with AI-based patent applications? For instance, which non-U.S. company has accumulated the most patent families?
Karthik: The single most striking development in the update is that the largest GenAI patent owner in the world is now a non-U.S. company that is not a traditional software or internet business: SoftBank, of Japan. SoftBank holds nearly 3,000 published patent families, virtually all of which appeared in 2025. The report attributes this surge to SoftBank's broad pivot toward AI — heavy investment in computing infrastructure, a central role in large-scale data-center initiatives, its position as a major financial backer of leading AI developers and the development of its own Japanese-language model. What makes SoftBank interesting is that it represents a new kind of GenAI patent owner: a telecommunications and investment conglomerate pursuing a vertically integrated strategy that spans infrastructure, models and applications, rather than a pure-play AI company.
Beyond SoftBank, China remains the center of gravity. Chinese inventors published more GenAI patent families in 2024 and 2025 alone than in the entire preceding decade, and the Chinese leaders from the 2024 report — Tencent, Ping An and Baidu — remain among the largest cumulative owners. SoftBank's rise also vaulted Japan past the Republic of Korea into third place among inventor countries, which illustrates a point worth keeping in mind when reading these rankings: The strategic decision of a single large filer can reshape an entire country's standing in a single year.
Frank: The WIPO statistics for China illustrate the importance of scale in driving GenAI technology. Significant drivers of AI growth include the scale of access to data for developing models, the infrastructure including data centers and the regulatory framework underlying the AI ecosystem. That said, scale is not the only driving force, so we will likely not see a progression to a winner-take-all market based on regional deployment of the technology.
WT: Besides the patent holding companies that you mention, how are some of the smaller companies from European nations, for instance, progressing?
Karthik: Europe's presence is more modest than that of China, the United States or Japan, but it is growing, and its character is telling. The most prominent European corporate entrant in the top ranks is Bosch, the German industrial and automotive supplier. Bosch's appearance is representative of how European industry is approaching GenAI: not by trying to build foundational models to rival the large U.S. and Chinese developers, but by applying generative techniques within its existing engineering, manufacturing and mobility businesses. This echoes the pattern we described earlier — established industrial companies protecting the use of AI in their own products and processes rather than the underlying models.
At the country level, Germany has now clearly established itself as the leading European location for GenAI invention, overtaking the United Kingdom, which had held that position in the 2024 report. Both countries have shown strong recent growth, but Germany's larger base and faster acceleration have widened the gap. Switzerland and Canada are also worth watching; both posted very high recent growth rates, which suggests that GenAI innovation, while still concentrated, is beginning to spread across a broader set of countries.
WT: What predictions do you have for the future of AI technologies, or technologies in general, based on the information in WIPO's report?
Karthik: A few things seem likely to us. First, the technological center of GenAI will keep moving toward large language models and, increasingly, reasoning-oriented and agentic systems — models that can plan and carry out multistep tasks with limited human oversight. The report already sees these beginning to surface in patent filings, and we expect that to accelerate. Alongside them, we expect a great deal of activity around efficiency — making models cheaper to train and to run. Much of the competitive contest, and much of the patenting, will be about doing more with less compute.
Second, the range of industries filing GenAI patents will keep broadening. As the technology works its way into healthcare, manufacturing, finance and energy, we expect the cast of applicants to look less like a list of software companies and more like a cross-section of the whole economy.
Third — and this is the note of caution we would sound for anyone reading the headline numbers — the published patent data tell only part of the story. Today, international patent families, meaning inventions that applicants choose to pursue in more than one country, still account for only a small share of GenAI filings, which tells us that much of this activity remains early-stage and domestic. As business models mature, we would expect more of these inventions to be extended internationally.
At the same time, a growing share of the most valuable AI innovation is being kept as trade secrets rather than patented, and that will never show up in a report like this one. So, while the trajectory is unmistakably steep, we would caution clients, investors and policymakers alike against treating patent counts as a complete measure of who is ahead. The most important developments may be the ones that are not being published at all.
Frank: AI will become more of a mindset than just a technology. Innovators looking at all use cases will by default ask first where AI fits in the product instead of whether it should be used. We are already seeing, for example, investors turning a cold shoulder to pitches that do not include an AI angle.
The trends discussed in the WIPO report are expected to continue so long as there is no disruptive force in this space. While most think about technical developments as being that disruptive force, we also are keeping a sharp eye on the law as it evolves to keep up with the technology. There is the potential for courts looking at issues of first impression presented by AI to make decisions that could chill the drive for patent protection.
Decisions, for example, calling into question whether AI systems that replicate human activity are abstract under the Alice framework [established in Alice Corp. v. CLS Bank International, 573 U.S. 208 (2014)], decisions raising the bar for determining whether AI inventions sufficiently advance the state of the art, or decisions limiting the value of AI patents by altering the damages calculus are all examples of the types of legal disruption that could change the trajectory of AI IP protection.
In general, when it comes to protection and enforcement, AI innovation highlights the importance of an IP triage process that has always existed, but it now takes on a new significance. In the AI IP triage process, a company must decide what is the most appropriate form of protection, often having to choose between seeking patent protection or maintaining the invention as a trade secret.
When choosing the patent path, the applicant has several hurdles. It must convince the examining patent office that the subject matter of the application is appropriate for patenting. This can be a challenge because many use cases for AI replicate human activity. The invention also has to be described in sufficient detail to permit the claims to be practiced, and the claimed invention has to advance the state of the art. Accomplishing all of those goals is far from straightforward given the evolving nature of AI technology and the case law interpreting statutes written well before GenAI was developed.
The AI inventor's dilemma: Do we file a patent application that discloses the invention and all of the company's "secret sauce." The application will be published and available for inspection by competitors worldwide in 18 months, and if the applicants do not succeed in overcoming the many complex hurdles for obtaining a patent, competitors can potentially practice what has been published without recourse. So, a company runs the risk of giving away its innovation by seeking patent protection. The other factor is that even if a company obtains patent protection, the half-life of the patented technology could be such that the technology is obsolete by the time the patent issues.
In view of all of this risk presented by the patent path, clients are increasingly following the trade-secret protection path for AI innovation. This is one trend that the WIPO statistics do not track because WIPO counts public patent application filings, but trade secrets are by their nature secret.
