Accelerating GenAI patenting activity
While the post-2017 growth in GenAI patenting documented in the 2024 WIPO Patent Landscape Report was already remarkable – averaging about 45% annually in the years following the introduction of the transformer architecture (2017–2023) – the latest data reveal a further acceleration. GenAI inventions (measured in terms of published patent families) roughly doubled between 2024 and 2025 (from 18,862 in 2024 to 37,808 in 2025), a growth rate that far surpasses the long-term average of 43% annually since 2014. The 2024 report anticipated this trend, noting that the surge in public interest and scientific research following the launch of ChatGPT in late 2022 would translate into increased patent activity with a time lag, given the typical 18-month lag between patent filing and publication. The 2024 and 2025 figures confirm this: the wave of GenAI research and development triggered by the release of ChatGPT in late 2022 and the broader commercial interest in LLMs is now fully materializing in the patent data.
![[Fig 1] GenAI patent family publications have surged, with 2024 and 2025 data confirming the accelerating post-ChatGPT innovation wave (Column Chart)](https://datawrapper.dwcdn.net/jbnQO/full.png)
A look at international patent families (IPFs) – patents that have been published in two or more jurisdictions – provides an additional perspective. IPFs are a useful indicator of the patents that applicants consider most commercially valuable, as the decision to bear the cost of filing in multiple countries typically reflects a strong intention to seek commercialization. The number of IPFs in GenAI has grown from 1,931 in 2023 to 3,297 in 2025. However, IPFs remain a small fraction of the total number of GenAI patent families, accounting for approximately 9% of all GenAI patent family publications in 2025. This gap is partly explained by the geographic composition of the published GenAI patent families, as discussed later. China accounts for the majority of global GenAI patent publications, but Chinese applicants typically file most of their patents domestically. The relatively young age of many GenAI patent portfolios is also a factor: applicants often file first in their domestic market and then include additional jurisdictions at a later stage, meaning that recently published patents may become IPFs as they mature.
GenAI is a growing share of all AI patenting activity
While GenAI remains a subset of the very dynamic broader AI patent landscape, its relative importance has continued to increase. In 2025, GenAI patent family publications accounted for approximately 8.7% of the total of more than 435,000 AI patent family publications, up from 6.1% in 2023 and 4.2% in 2017. As GenAI technologies – particularly LLMs, diffusion models and multimodal systems – continue to find their way into an expanding range of products and applications, the GenAI share of total AI patenting is expected to continue its upward trajectory.
Top patent owners
The most striking change in the ranking of top GenAI patent owners is the emergence of the Japanese company SoftBank as the single largest patent holder. It had 2,985 patent families published between 2014 and 2025 – virtually all of which were filed in 2023 and published in 2025. This extraordinary growth aligns with SoftBank’s broad strategic pivot toward AI. The company has invested heavily in AI infrastructure – including constructing large-scale graphics processing unit (GPU) computing platforms in Japan and taking a leading role in the Stargate data center initiative in the United States
The Chinese companies that dominated the 2024 report rankings – Tencent, Ping An and Baidu – remain among the largest cumulative patent holders. Several organizations have climbed significantly in the rankings compared with the 2024 report. State Grid Corporation of China – the world’s largest utility company by number of customers served and workforce
Among companies based in the United States, Alphabet (Google) (1,083 patent families) has strengthened its position, becoming the largest US-based GenAI patent owner and the leading US applicant in terms of recent filings (640 in 2024 or 2025). Microsoft (865) and IBM (821) follow in second and third places, although IBM’s recent filing rate (220 in 2024 or 2025) is comparatively lower. Nvidia (497 patent families, with 333 in 2024 or 2025) is a notable new entrant in the top 25, reflecting its expanding role beyond GPU hardware into AI software, frameworks and model architectures. Adobe (546) also continues to build its GenAI portfolio.
New entrants in the ranking also include: China Southern Power Grid (622 patent families); Inspur Group (486), a major Chinese server and cloud computing provider; and Bosch (368), which represents the growing engagement of European industrial companies in GenAI patenting. Ant Group (353), the fintech affiliate of Alibaba, also appears in the top 25 GenAI patent owners.
The 2024 report noted that OpenAI – the company behind ChatGPT – had only just begun to file patents. Since then, OpenAI has continued to build its patent portfolio, although it remains modest in comparison with other leading GenAI companies. As of late 2025, OpenAI held 35 patent filings globally, a tiny fraction compared with the thousands of published patents held by companies like Alphabet (Google), Microsoft or Tencent. Its patents focus on specific product-level innovations – multimodal interfaces, code generation, image generation and text editing – rather than foundational model architectures. OpenAI’s approach suggests a continued reliance on trade secrets and speed of execution as its primary competitive advantages, with patents serving a complementary role.
Top inventor locations
China remains firmly at the forefront of global patenting activity in GenAI. China’s output has continued to accelerate, with over 43,000 patent families published in 2024 and 2025 alone, more than its entire cumulative output over the preceding 10 years (2014 to 2023). China’s compound annual growth rate (CAGR) for patent families is 64% between 2023 and 2025, which demonstrates that its dominance is not merely a legacy of early filing volume but reflects sustained and intensifying research activity.
The United States remains the second most important research location. The number of US patent families rose from about 1,185 in 2023 to over 4,380 in 2025, suggesting that the wave of commercial GenAI development led by US companies is now translating into a significant acceleration in patenting activity.
A remarkable shift seen in the updated data is the rise of Japan, which has moved from fourth to third place among inventor locations, overtaking the Republic of Korea. Japan’s patent output jumped from just 398 patent families in 2023 to 3,835 in 2025 – showing a CAGR of 210%, by far the highest among the top inventor locations. This dramatic increase is largely attributable to the patenting activity of SoftBank, which published nearly 3,000 GenAI patent families in 2024 and 2025, as discussed previously. Without SoftBank’s contribution, Japan’s growth would be significantly more modest.
![[Fig 4] China continues to dominate the GenAI patent landscape, but patent growth has been more dynamic in the United States, Japan, Germany, Canada and Switzerland recently. Seven of the top 10 inventor locations have seen more GenAI patent families published in the last two years than their entire cumulative output over the preceding ten years. (Grouped column chart)](https://datawrapper.dwcdn.net/g0bO8/full.png)
The Republic of Korea has dropped from third to fourth place, with a more moderate growth rate in published patent families (21% CAGR between 2023 and 2025). India (fifth) continues to show strong growth (63% CAGR), consistent with the high growth rates already noted in the 2024 report.
Among European locations, Germany (sixth place) has now clearly overtaken the United Kingdom (eighth place) and has established itself as the leading European inventor location for GenAI. Both countries have shown strong recent growth rates (85% CAGR for Germany, 69% for the UK), but Germany’s larger base and faster acceleration have widened the gap observed in the 2024 report.
Overall, the geographic concentration of GenAI patenting remains high. The top four inventor locations alone (China, the United States, Japan and the Republic of Korea) account for the vast majority of global GenAI patent family publications. However, it should be noted that some patent families list inventors from more than one country, meaning that country-level figures contain a degree of overlap. The updated data also reveal that the growth in GenAI patenting is becoming more broadly distributed, with several key countries, including the United States, Japan, Germany, Canada and Switzerland, posting CAGR figures above 80%, narrowing the gap with China in relative terms even as China continues to lead in absolute volume.
Main GenAI models
As in the 2024 report, all identified GenAI patent families were assigned to five different model types based on information from patent abstracts, claims or titles. However, not all GenAI patent families fit into a specific model type, as some patents focus on describing the use case rather than the underlying model architecture.
A significant change in the updated data is that LLMs have overtaken generative adversarial networks (GANs) as the largest GenAI model category by patent volume. Between 2014 and 2025, approximately 20,900 patent families were published in the LLM category, compared with about 18,800 for GANs. This reversal was already foreshadowed in the 2024 report, which noted the rapid growth of LLM patents and the slowing momentum of GAN filings, but the speed at which LLMs have moved ahead is noteworthy. In 2023, LLM patent families (881) were still well below GANs (2,370). By 2025, LLMs had increased to over 14,100 patent families – nearly three times the number of GAN patents published in the same year (5,245).
![[Fig 5] Large Language Models (LLMs) are now the central model type driving the GenAI wave, overtaking Generative Adversarial Networks (GANs) (Line chart)](https://datawrapper.dwcdn.net/dGtQt/full.png)
This shift reflects the broader technological transition in the GenAI field. While GANs were the dominant generative model for much of the past decade, particularly for image synthesis and data augmentation, the transformer-based architectures underpinning LLMs have become the central technology driving the current GenAI wave. The commercial success of LLM-based products, such as ChatGPT, Gemini, Claude and others, has clearly translated into increased patenting activity. GANs remain relevant in specific applications of AI, such as image enhancement, synthetic data generation and real-time inference tasks, but they are no longer at the technological forefront of GenAI research and development.
Diffusion models have also experienced strong growth, rising from 441 patent families in 2023 to nearly 4,000 in 2025. These models, which underpin widely used image and video generation tools, are now the third largest model category by patent volume (approximately 6,500 patent families in total), having overtaken variational autoencoders (VAEs). VAEs themselves have also grown substantially – from 562 patent families in 2023 to over 2,370 in 2025.
Autoregressive models remain the smallest category, with about 1,280 patent families in total between 2014 and 2025. As noted in the 2024 report, there is inherent overlap between autoregressive models and LLMs, since LLMs are by definition autoregressive. The relatively weak growth in the autoregressive category (from 126 in 2023 to 557 in 2025) is likely to reflect a terminological shift: as LLMs have become the dominant framing in both academic research and industry, patent applicants appear to be using the term “large language model” rather than “autoregressive model” in their filings.
Overall, the data show that the patent landscape has moved decisively from a GAN-dominated era to one led by LLMs and, increasingly, diffusion models. This mirrors the broader technological evolution of GenAI – from a field centered on adversarial image generation to one defined by large-scale language and multimodal models with wider application potential.
Main GenAI modes
GenAI models can use many different types, or modes, of input and output data, such as text, image/video, speech/voice/music and others.
A notable development in the updated data for 2024 and 2025 is the rapid rise of text as a mode. Text-related patent families grew from approximately 3,400 in 2023 to nearly 11,800 in 2025, more than tripling in just two years. This surge closely mirrors the strong growth in LLM-related patents and reflects the central role that text generation, processing and understanding now plays in the GenAI ecosystem.
The speech/voice/music category has fallen behind in terms of patent growth dynamics, with patent families using this mode declining in 2024 before recovering to 3,538 in 2025. This relative slowdown is seemingly at odds with the active commercial development of voice and audio AI technologies in 2024 and 2025, including real-time speech-to-speech models, AI voice cloning and AI music generation tools. One possible explanation is a classification effect: as modern voice and audio systems are increasingly built on top of multimodal LLMs, rather than using dedicated speech architectures, related patents may be classified under text or other modes rather than speech/voice/music.
Software/code has emerged as one of the fastest-growing modes in relative terms. Patent families in this category rose from 339 in 2023 to 1,616 in 2025, reflecting the growing importance of AI-assisted code generation tools and the integration of GenAI into software development workflows. While still a comparatively small category in absolute terms, its recent trajectory suggests that it will continue to gain prominence. 3D image modeling has also shown strong growth, rising from 908 patent families in 2023 to 2,484 in 2025. This growth aligns with increasing interest in GenAI applications for 3D content creation, virtual environments, gaming and product design.
The molecules/genes/proteins category – highlighted in the 2024 report as one of the fastest-growing modes – has shown more uneven growth in recent years, with 523 patent families in 2023, a dip to 397 in 2024, and a recovery to 766 in 2025. While this remains a relatively small category, it continues to reflect the growing application of GenAI in drug discovery, protein design and the life sciences more broadly.
![[Fig 6] Among the different GenAI modes, most patent families belong to the image, video category, but text and other modes are increasingly catching up (Line chart)](https://datawrapper.dwcdn.net/JqO25/full.png)