WIPO Report: Generative AI Patents Surge by Over 56,000 in Two Years

The World Intellectual Property Organization (WIPO) recently released its Generative AI Patent Trends Update report. The data show that between 2024 and 2025, the number of globally published generative AI patent families exceeded 56,000, surpassing the total for the entire previous decade. This marks a full-scale transition of the field from laboratory research to large-scale commercial deployment.

According to the report, in 2025 alone, approximately 37,800 generative AI patent families were published globally, nearly doubling the 18,900 published in 2024. The average annual growth rate far exceeds the long-term average of 43%. Meanwhile, the share of generative AI patents as a proportion of all AI patents rose from 6.1% in 2023 to 8.7% in 2025, indicating that technological innovation activity in this area is increasingly outpacing other AI subfields.

The ranking of patent holders has undergone dramatic shifts. Japan's SoftBank, with nearly 3,000 patent families, jumped to become the world's largest holder of generative AI patents, the vast majority of which were published in 2025. This reflects the company's extensive strategic investments in AI infrastructure, large language models, and data centres.

Among Chinese companies, Tencent, Ping An Insurance, and Baidu remain in the top ten in terms of cumulative holdings. New entrants to the top 25 include State Grid, NVIDIA, Inspur Group, and Bosch, reflecting that the energy, hardware manufacturing, and industrial sectors are accelerating their generative AI patent portfolios. On the U.S. side, Google (Alphabet) leads with 1,083 patents, making it the largest holder in that country, followed by Microsoft and IBM. OpenAI, by contrast, holds only 35 global patents to date, still relying primarily on trade secrets and first‑mover speed advantages.

By inventor location, China remains the undisputed global leader in absolute numbers, with over 43,000 publications in 2024–2025, exceeding its total for the previous decade and recording a compound annual growth rate of 64%. The United States ranks second, with approximately 4,380 publications during the same period.

The biggest "dark horse" is Japan, whose patent publications surged from 398 in 2023 to 3,835 in 2025, representing a compound annual growth rate of 210%, propelling it past South Korea to third place globally – largely attributable to SoftBank's large‑scale filings. In addition, countries such as Germany, Canada, and Switzerland all recorded growth rates exceeding 80%, showing that the global innovation landscape is becoming more diversified.

In terms of model types, large language models (LLMs) have completely overtaken generative adversarial networks (GANs) as the dominant technology. In 2025, LLM‑related patent publications exceeded 14,100, nearly three times that of GANs (5,245). Diffusion models also rose rapidly, taking third place with nearly 4,000 publications. This shift indicates that the technological focus of generative AI has moved away from early image‑based adversarial generation towards Transformer‑based text and multimodal generation.

With respect to application modes, image/video‑related patents are the most numerous, with over 13,800 in 2025, covering various technologies including GANs, diffusion models, and multimodal LLMs. However, text‑based applications are growing fastest, rising from approximately 3,400 in 2023 to nearly 11,800 in 2025, closely mirroring the LLM boom. Software/code, 3D modelling, and other directions also show robust growth, while speech/music categories have grown relatively more slowly, possibly because relevant technologies are being subsumed under multimodal LLM classifications.

The report concludes by emphasising that as multimodal systems, reasoning models, AI agents, and training‑inference efficiency optimisation become new hotspots, patent competition in generative AI will intensify. For policymakers, corporate decision‑makers, and IP practitioners, continuously monitoring this rapidly evolving patent landscape has become essential for grasping technological trends and competitive dynamics.

【Appendix】Patent Trends Update in GenAI