Generative artificial intelligence has moved into a different phase of patent activity.
Only a few years ago, much of the patent discussion around GenAI centred on generative adversarial networks, early language-generation techniques and relatively narrow applications. The position in 2026 is considerably different. Large language models (LLMs) are now an important patent category in their own right; diffusion-based technologies have expanded rapidly; GenAI is being incorporated into industrial and enterprise systems; and companies outside the conventional software sector are building patent portfolios around the technology.
The numbers illustrate the pace of that change. According to the World Intellectual Property Organization’s 2026 Patent Trends Update in GenAI, published GenAI patent families increased from approximately 14,000 in 2023 to more than 37,000 in 2025. More than 56,000 GenAI patent families were published during 2024 and 2025 alone more than the cumulative number published during the entire 2014–2023 period. GenAI also increased its share of overall AI patenting from 6.1% in 2023 to 8.7% in 2025.
For patent applicants and technology companies, however, filing volume is only part of the story. The more interesting development is what companies are now seeking to protect, where that activity is occurring, and how the underlying patent landscape is changing as GenAI moves into commercial deployment.
The GenAI patent surge accelerated after 2023
WIPO’s earlier 2024 Patent Landscape Report identified approximately 54,000 GenAI inventions during the decade ending in 2023. More than one-quarter of those had been published in 2023 alone. At that point, Tencent, Ping An Insurance and Baidu occupied the leading positions, while China was already substantially ahead of other inventor locations.
The updated figures through 2025 show that this was not a temporary filing spike.
Published GenAI patent families roughly doubled from 18,862 in 2024 to 37,808 in 2025. This acceleration corresponds with the period in which large language models moved from research environments into enterprise software, search, coding, productivity tools, healthcare, financial services, and other commercial applications.
Patent publication data necessarily carries a time lag because applications are generally not published immediately after filing. The 2025 publication figures should therefore not be read as a complete picture of inventions conceived or applications filed during 2025. They are better understood as evidence of the substantial R&D and patent-filing activity already underway in the preceding period.
That qualification is particularly important when analysing a technology developing as quickly as GenAI.
Read Also – How AI Is Changing Patent Examination and Prosecution in 2026?
China remains the largest source of GenAI patent activity
The geographic concentration of GenAI patenting remains striking.
China continues to rank first in GenAI patent publications, followed in WIPO’s 2025 data by the United States, Japan, the Republic of Korea and India. Germany, Canada, the United Kingdom, Israel and Switzerland complete the top ten inventor locations.
China-based inventors were responsible for more than 43,000 GenAI patent families published during 2024 and 2025. Six of the ten largest GenAI patent applicants are Chinese organizations.
The position should nevertheless be examined with some care. Patent volume is not the same as patent quality, commercial value, enforceability or technological leadership.
Stanford’s 2026 AI Index, looking at AI more broadly rather than GenAI alone, reports a more complex competitive picture: China leads in AI research publication volume, citations and patent grants, while the United States remains particularly strong in the development of notable AI models and in higher-impact patents.
For IP strategy, therefore, simply comparing national filing totals can be misleading. Patent-family quality, jurisdictions selected for protection, claim scope, citation impact, continuation strategy, commercial implementation and the importance of the underlying technology may matter considerably more than the raw number of applications.
The list of leading patent owners is changing
One of the most notable findings in WIPO’s 2026 update is the change in the applicant rankings.
The leading GenAI patent owners identified by WIPO are:
| Rank | Patent owner |
| 1 | SoftBank |
| 2 | Tencent Holdings |
| 3 | Ping An Insurance Group |
| 4 | Baidu |
| 5 | Chinese Academy of Sciences |
| 6 | State Grid Corporation of China |
| 7 | Alphabet/Google |
| 8 | Zhejiang University |
| 9 | Microsoft |
| 10 | IBM |
SoftBank’s movement rise to the top is particularly noteworthy. WIPO reports nearly 3,000 GenAI patent families associated with SoftBank, almost all published recently.
Among U.S.-based companies, Alphabet/Google has become the largest GenAI patent owner, with 1,083 patent families in WIPO’s updated dataset. Microsoft follows with 865 and IBM with 821. Nvidia has also entered the top 25, with 497 patent families, 333 of which were published in 2024 or 2025.
The presence of Nvidia is particularly indicative of how the GenAI patent landscape is broadening. Patent competition is no longer confined to companies developing consumer-facing generative applications. The relevant IP can extend across model architectures, computing infrastructure, training and inference techniques, software frameworks, data processing and implementation of AI within particular technical systems.
GenAI patenting is expanding beyond traditional technology companies
Another development deserves attention.
State Grid Corporation of China now holds more than 1,100 GenAI patent families, according to WIPO. Its portfolio reflects applications of GenAI in areas including grid optimization, predictive maintenance and infrastructure planning. Bosch has also appeared among notable newer entrants to the leading applicant group.
This changes the character of the patent landscape.
The first generation of GenAI patent portfolios was heavily associated with internet companies, software businesses and specialist AI research organizations. The current landscape increasingly includes companies operating in energy, infrastructure, telecommunications, industrial systems, finance and other sectors.
That suggests that an increasing share of future patent disputes and licensing questions may concern the application of generative AI within particular industries, rather than only the foundational models themselves.
For companies conducting patentability searches or freedom-to-operate assessments, this distinction can become important. A business deploying GenAI in an industrial product may need to consider not only patents directed to the underlying AI architecture, but also patents covering how generative techniques interact with the particular technical system in which they are implemented.
Large language models have overtaken GANs
The technological composition of the patent landscape has also changed substantially.
Generative adversarial networks, or GANs, were one of the defining architectures of the earlier GenAI patent landscape. They remain important, but they are no longer the largest model category.
WIPO reports approximately 14,100 LLM-related patent-family publications in 2025, compared with approximately 5,200 relating to GANs. Large language models have therefore overtaken GANs by a considerable margin.
Diffusion models have moved into third position. Variational autoencoders and autoregressive models remain important additional categories.
This shift reflects what has occurred technically and commercially. GenAI development has moved substantially toward language-based systems capable of generation, reasoning-related tasks, retrieval interaction, software assistance and multimodal processing.
The patent implications go beyond simply claiming an LLM.
As foundational model techniques become increasingly documented in patents, research papers and open-source implementations, obtaining meaningful patent protection may depend on identifying a more specific technical contribution: improvements to training, inference, model architecture, memory, retrieval, compression, deployment, security, hardware interaction or a particular technical application.
A broad description of “using AI to generate an output” will not necessarily provide a strong patent position.
Also Read – Patent Ownership and Assignment in India: What the IIT Bombay Case Tells Inventors and Research Institutions
Diffusion models represent another important area of patent activity
Diffusion models have become particularly significant in image and video generation and are increasingly relevant to other data types.
WIPO’s earlier patent-landscape work already identified rapid growth in this category and found Zhejiang University particularly active in diffusion-model patenting. China has maintained a substantial lead in patent-family activity relating to these models.
From a patent perspective, diffusion technology can give rise to protection at several levels: model architecture, training methodology, noise scheduling, conditioning mechanisms, computational optimization and technical applications built around generated output.
As with LLMs, however, the rapidly expanding prior-art base changes the prosecution environment.
A patent application filed today enters a field containing not only rapidly growing patent literature, but also extensive academic publications, conference papers, repositories, technical disclosures and open-source implementations. Establishing novelty and inventive step may consequently require a much more precise identification of the technical contribution than would have been necessary during an earlier stage of the technology.
Image and video still represent the largest GenAI mode
Despite the extraordinary attention given to text-based systems, image and video remain the largest GenAI category when WIPO’s data is considered across the period from 2014 through 2025.
Approximately 40,000 patent families were published during that period related to image/video GenAI, with more than 13,800 published in 2025 alone.
Text, however, is closing the gap rapidly. WIPO reports that text-related GenAI patent filings more than tripled within two years. Software/code and 3D modelling are also among the faster-growing areas.
This is important because GenAI is increasingly becoming multimodal.
A modern AI system may process combinations of text, images, video, speech, audio and structured data rather than operating within a single modality. Patent specifications drafted around only one present implementation may therefore fail to capture commercially relevant variations of the underlying invention.
Where the disclosure supports it, applicants may need to consider carefully whether the inventive concept extends across multiple input and output modes and how those alternatives should be reflected in the specification and claims.
Life sciences may become one of the more consequential GenAI patent areas
Generative AI patenting is also moving beyond conventional content generation.
WIPO’s 2024 analysis identified GenAI inventions involving molecules, genes and proteins as a comparatively smaller but rapidly growing category. It recorded 1,494 such inventions from 2014 through 2023 and an average annual growth rate of 78% over the preceding five years.
This area is particularly interesting from an IP perspective because the patent questions can extend well beyond protection for an AI model.
Generative systems may be used to propose molecular structures, identify candidate compounds, design proteins or assist other stages of scientific discovery. The resulting IP strategy may therefore involve questions concerning patent protection for the computational method, the generated candidate, downstream compositions or products, methods of use and associated experimental developments.
Inventorship can also become more difficult where AI contributes materially to the process leading to an invention. Patent systems generally continue to require human inventors, but determining the human contribution to AI-assisted research may become increasingly important as generative systems participate more directly in R&D.
The patent landscape is becoming more commercially significant and more difficult to navigate
The sheer growth in patent filings creates a strategic problem of its own.
A company developing GenAI technology now operates within a much denser prior-art and patent environment than it did even three or four years ago. Relevant rights may be distributed among software companies, universities, infrastructure providers, financial institutions, electronics manufacturers and specialist AI businesses.
The result is that patentability and freedom to operate become increasingly distinct exercises.
A company may obtain a patent for an improvement to an AI system while still needing to consider third-party patents relevant to commercial implementation. Conversely, identifying a large number of patents within a technology area does not establish that those patents cover a particular product or activity. Claim construction, legal status, territorial coverage, priority dates and the specific technical implementation all require examination.
For businesses entering crowded areas such as LLM infrastructure, multimodal AI, AI-assisted coding or industrial GenAI, patent landscaping can therefore serve a broader purpose than identifying competitors. It can reveal technological concentration, white spaces, emerging filing patterns and potential areas requiring closer freedom-to-operate analysis.
Patent quantity should not be mistaken for patent strength
The extraordinary growth figures in GenAI require one final qualification.
A published patent family is evidence of patenting activity. It is not evidence that the invention will ultimately produce a granted, enforceable or commercially important patent.
Applications may be refused, withdrawn or substantially narrowed during prosecution. Granted claims may cover only a limited implementation. A patent family may have protection in commercially important jurisdictions or only in a relatively narrow geographic market.
The competitive significance of a GenAI portfolio therefore cannot be assessed reliably from portfolio size alone.
A more meaningful analysis would consider claim scope, patent-family coverage, priority position, prosecution status, remaining patent term, citation patterns, technical relevance and correspondence between the claims and commercially implemented technology.
This is particularly relevant in GenAI because the technology is moving faster than the patent publication cycle. A portfolio that appears numerically dominant today may reflect an earlier technological direction, while commercially significant inventions filed more recently may not yet be publicly visible.
What the 2026 landscape means for patent strategy?
The GenAI patent landscape has reached a stage where companies should think beyond simply filing applications around every AI-related development.
The prior-art environment is becoming crowded. Patent offices have access to increasingly sophisticated search technologies. Academic and open-source disclosure is expanding quickly. At the same time, competitors are filing across both foundational AI technologies and industry-specific implementations.
For patent applicants, this places greater importance on identifying the actual technical contribution before drafting begins.
For established companies, portfolio review should consider whether existing claims remain aligned with the direction in which GenAI technology is developing.
For businesses adopting third-party foundation models, the fact that the underlying model is externally supplied should not automatically be taken to mean that patent risk disappears. The company’s own implementation, integration and technical application may still sit within a wider patent landscape.
And for organizations assessing competitors, raw filing numbers should remain the starting point rather than the conclusion.
The 2026 data ultimately shows something more significant than rapid patent growth. Generative AI is moving from a concentrated field of AI research into a technology layer being incorporated across industries. Patent activity is following that movement.
For IP professionals, R&D teams and technology companies, the question is changing. It is no longer simply: who is patenting generative AI?
The more important questions are becoming: what part of the GenAI technology stack is being protected, how strong and geographically relevant are those rights, and where is the next concentration of patent activity likely to emerge?
Those questions will increasingly shape patent drafting, prosecution, portfolio development, competitive intelligence and freedom-to-operate strategy in the years ahead.
This article is intended for general information concerning patent trends in generative artificial intelligence. Patent statistics and rankings are based on the cited datasets and should not be interpreted as assessments of patent validity, enforceability, portfolio quality or commercial value. Specific patentability or freedom-to-operate questions require analysis of the relevant claims, prior art, jurisdictions and factual circumstances.
AnalystIP Legal Services Pvt. Ltd.
Patent Drafting | Patent Analytics & Landscape Studies | Patentability Searches | Freedom-to-Operate Analysis | Patent Prosecution | IP Research & Advisory
Sources and References
- World Intellectual Property Organization (WIPO), Patent Trends Update in GenAI (2026) — the principal source for the updated 2025 patent-family data, applicant rankings, geographical trends and technology categories.
WIPO – Patent Trends Update in GenAI - WIPO, “Top Generative AI Patent Trends in 2025” — updated statistics on GenAI inventions, leading patent owners, inventor locations, model categories and data modalities.
WIPO – Top Generative AI Patent Trends in 2025 - WIPO, “GenAI Innovation Soaring, With Patent Activity Nearly Tripling in Two Years,” 14 July 2026 — WIPO’s release accompanying the 2026 update, including recent growth and geographic concentration data.
WIPO – GenAI Innovation Soaring - WIPO, Patent Landscape Report: Generative Artificial Intelligence (2024) — underlying landscape study covering the earlier 2014–2023 period and providing data on applicants, models, modalities and application fields.
WIPO – Generative AI Patent Landscape Report - WIPO, Patent Trends in GenAI Models — detailed analysis of LLMs, GANs, diffusion models, VAEs and autoregressive models, including applicant and geographic patterns.
WIPO – Patent Trends in GenAI Models - WIPO, Patent Trends in GenAI Applications — analysis of patent ownership across software, business solutions, life sciences, transportation, security, telecommunications and other application areas.
WIPO – GenAI Application Patent Trends - Stanford Institute for Human-Centered Artificial Intelligence, AI Index Report 2026 – Research and Development — broader AI research, model-development and patent trends, including comparative activity in China and the United States.
Stanford HAI – 2026 AI Index, Research and Development