IP News
Generative AI Patent Families Triple in Two Years as China and SoftBank Reshape the Landscape
•
10 min read
•162 views
//
Global generative AI patent publications more than tripled in 2024–2025, with the two-year total exceeding the cumulative volume of the preceding decade. China remains the dominant source of these patent families, but Japan’s SoftBank Group has surged to the top entity spot, and the United States is growing at a faster compound annual rate. The data, published by WIPO on 15 July 2026, signals an intensifying patent thicket that demands immediate portfolio review by corporate IP leaders.
On 15 July 2026, the World Intellectual Property Organization (WIPO) published a patent landscape report on generative artificial intelligence. The report shows that 56,000 generative AI patent families were published globally in 2024 and 2025 combined, a volume that exceeds all generative AI patent families published from 2013 through 2023. In 2025 alone, 37,800 new patent families were published, up from 14,000 in 2013.
China accounted for more than 43,000 of the patent families published during 2024–2025. The United States published 4,380 patent families in 2025, but its compound annual growth rate (CAGR) from 2023 to 2025 reached 92%, outpacing China’s 64% CAGR over the same period. Japan recorded a 210% CAGR, driven almost entirely by SoftBank Group.
SoftBank claimed the top entity spot with nearly 3,000 published patent families since 2013, nearly all of which were published in 2024–2025. Chinese entities occupied four of the top five positions: Tencent Holdings (2,702 patent families), Ping An Insurance Group (2,240), Baidu (1,902), and the Chinese Academy of Sciences. Six of the top ten patenting entities are based in China. U.S. companies in the top ten include Alphabet, Microsoft, and IBM.
The report also documents a shift in model architecture. Large language models (LLMs) outnumbered generative adversarial networks (GANs) nearly three-to-one in their share of 2025 generative AI patent families. Image and video remained the primary generative mode, appearing in roughly 40,000 patent families published in 2024–2025, though text-mode patents grew sharply from 3,400 families in 2023 to nearly 11,800 in 2025.
WIPO’s 2026 update builds on its 2024 generative AI patent landscape report, which covered 2014–2023 and identified 54,000 generative AI patent families worldwide. That earlier report showed China with more than 38,000 patent families, six times the U.S. total of approximately 6,300. The new data confirms that the acceleration first noted in 2024 has intensified.
Patent family counts measure published filings, not grants, and a single invention may generate multiple family members across jurisdictions. The figures therefore reflect filing activity and disclosure volume rather than enforceable rights. Chinese patenting incentives—including cash subsidies, High and New Tech Enterprise (HTNE) tax rates, and listing requirements for the Shanghai Stock Exchange Science and Technology Innovation Board—likely inflate filing volumes relative to commercial deployment. The U.S. and China are nearly tied in scientific paper output on generative AI, but U.S. papers receive significantly more citations, suggesting a gap between patent quantity and research influence.
For Korean practitioners, the Republic of Korea ranked third globally in the 2014–2023 dataset with 4,155 generative AI patent families, behind China and the U.S. but ahead of Japan and India. The 2026 update does not break out Korea-specific figures, but the trajectory implies continued strong filing activity by Korean entities, including Samsung Electronics, which appeared in the top ten in the earlier report.
The density of published patent families creates a material freedom-to-operate burden. A company developing text-to-image, LLM-based, or multimodal generative AI products must now clear a landscape where more than 56,000 patent families were published in two years. The risk is not theoretical: Chinese entities hold large portfolios across image generation, speech synthesis, and digital-human technologies, and SoftBank’s sudden accumulation of nearly 3,000 families in a two-year window suggests an aggressive assertion strategy may follow.
For Korean filers, the development cuts both ways. Samsung, LG, and Korean AI startups are patentees seeking protection in the U.S., China, and Europe; a crowded field raises the bar for obtaining broad claims. But those same companies are also potential defendants or licensees facing Chinese and Japanese portfolios. A Korean company challenging a Chinese-origin patent in a post-grant opposition or invalidation proceeding benefits from the fact that high-volume filing programs often produce claims with narrower technical contribution, making them more vulnerable to prior-art attacks. Portfolio managers should map their own generative AI filings against the top entities identified by WIPO and flag overlapping claim scope.
Drafting attorneys should adjust claim strategy now. Because LLM-related patent families are growing faster than GAN-related families, claims directed to text-generation methods, prompt engineering, and retrieval-augmented generation will face increasingly dense prior art. Draft claims with fallback positions tied to specific training data structures, fine-tuning pipelines, or hardware-software integration points that are less likely to appear in the high-volume portfolios of the top Chinese filers. When prosecuting before KIPO, remember that Korea’s added-matter bar under Article 47(2) of the Patent Act is strict; amendments that broaden or shift the technical contribution during prosecution will be rejected, so front-load the specification with multiple concrete implementation examples that support narrow fallback claims.
WIPO is likely to update this dataset biennially. The next report will show whether SoftBank’s filing surge was a one-time portfolio build or the start of sustained activity. It will also reveal whether U.S. growth rates continue to outpace China’s, and whether Korean entities regain a top-ten presence.
In the interim, IP owners should run a patent landscape search limited to 2024–2025 published families in their sub-domain, using the WIPO taxonomy as a filter. Identify the top three entities whose portfolios overlap with your own product roadmap. For each, pull the five most-cited or most geographically extended families and assess whether a post-grant opposition or invalidation search is warranted. If your organization is a Korean challenger, prioritize Chinese-origin families that have been filed only in China and the U.S.; these may lack the procedural history that strengthens presumption of validity in other jurisdictions. If your organization is a patentee, review pending applications for claims that read on SoftBank’s published families and consider accelerating examination where possible to secure issuance before the landscape grows denser.