Generative AI is no longer only a software and model story. Patent data points to a broader shift into robotics, autonomous machines, manufacturing, and other physical systems.
According to the World Intellectual Property Organization (WIPO), more than 56,000 GenAI patent families were published in 2024 and 2025, exceeding the total published during 2014–2023. Published GenAI patent families rose from about 14,000 in 2023 to more than 37,800 in 2025.
China is at the center of this expansion. Chinese inventors accounted for more than 43,000 GenAI patent families published in 2024 and 2025, while China’s GenAI patent-family output grew at a 64% CAGR between 2023 and 2025.
But the more important IP story may be what happens next.
As generative models become embedded in machines that can perceive, reason and act, patent competition is likely to move beyond models and content generation toward physical AI: technologies that allow robots, autonomous vehicles, drones and industrial systems to operate in the real world.
GenAI patenting has entered a new phase
WIPO’s latest data shows how quickly the landscape has changed.
| Indicator | Latest WIPO data |
| GenAI patent families, 2023 | ~14,000 |
| GenAI patent families, 2025 | >37,800 |
| GenAI families published in 2024–2025 | >56,000 |
| GenAI share of all AI patent families | 8.7% |
| China-based GenAI families, 2024–2025 | >43,000 |
| China GenAI CAGR, 2023–2025 | 64% |
| US GenAI CAGR, 2023–2025 | 92% |
| Japan GenAI CAGR, 2023–2025 | 210% |
China has a large lead in absolute GenAI patenting volume, but growth is accelerating elsewhere. Japan’s GenAI patent families, for example, increased from about 398 in 2023 to 3,835 in 2025.
Patent volume, however, does not equal patent strength.
International GenAI patent families increased from 1,931 in 2023 to 3,297 in 2025, but represented only about 9% of GenAI patent-family publications in 2025. The distinction between filing activity and commercially significant international protection is therefore important.
China is leading the volume race
WIPO’s 2024 GenAI patent landscape identified more than 54,000 GenAI inventions globally during 2014–2023, with more than 38,000 originating from China, about six times the US total.
The latest update shows continued acceleration. China-based inventors produced more than 43,000 GenAI patent families in 2024 and 2025 alone.
China also accounted for 73,718 PCT international patent applications in 2025, ahead of the United States at 52,617 and Japan at 47,922.
However, China’s large GenAI patent-family volume does not automatically translate into the same level of international protection. A large share of Chinese filings remains domestic. For competitors, those filings can still matter because published applications may become relevant to later prior-art analysis, depending on the jurisdiction.
The technology itself is changing
The patent data also shows a shift in what inventors are protecting.
In 2025, WIPO identified about 14,100 LLM-related patent-family publications, compared with about 5,200 involving GANs. Diffusion models ranked third.
Image and video remained the largest GenAI data-mode category, with roughly 40,000 patent families published between 2014 and 2025. Text, software/code and 3D modelling are also expanding.
This matters because physical AI requires more than content generation.
A machine operating in the real world must combine:
Perception → interpretation → planning → decision-making → control → physical action
That creates a much larger IP surface than a conventional chatbot.
This is where GenAI meets physical AI
Physical AI refers broadly to AI systems that can perceive and act in the physical world. It includes autonomous vehicles, industrial robots, warehouse systems, drones, agricultural machines, surgical robots and humanoid robots.
Existing autonomous systems already rely on sensing, perception, communication, AI-based data processing and machine control. Generative AI adds another layer.
A foundation model could help a machine interpret its environment, understand natural-language instructions, generate plans, adapt to new situations or interact with humans.
This creates potential patent opportunities at the interface between AI models and physical systems.
Where the next patent clusters could form
| Physical-AI layer | Emerging IP questions |
| Perception | Multimodal sensing, vision-language, sensor fusion |
| World modelling | Representation of physical environments and objects |
| Planning | AI-generated task and motion planning |
| Control | Translating model outputs into safe physical actions |
| Human-machine interaction | Natural-language programming and instruction |
| Robotics | General-purpose and task-specific foundation models |
| Autonomous vehicles | Generative planning, simulation and decision systems |
| Industrial systems | AI-driven production, inspection and maintenance |
| Drones | Adaptive navigation and environment-aware control |
| Edge AI | Models operating under power and latency constraints |
Many of these technologies do not require a new foundation model. A company may instead develop a novel way of using an existing model within a particular physical system.
The patent race is spreading beyond AI companies
WIPO’s leading GenAI patent applicants include SoftBank, Tencent, Ping An, Baidu, the Chinese Academy of Sciences, State Grid Corporation of China, Alphabet, Zhejiang University, Microsoft and IBM.
The list shows that GenAI patenting is spreading into infrastructure, finance and industrial applications.
State Grid, for example, has GenAI patent activity related to energy-grid optimisation, predictive maintenance and infrastructure planning. Industrial companies are also applying AI to manufacturing.
The shift is important: companies are protecting not only how AI works, but how AI is applied to a technical or industrial problem.
In physical AI, a robotics company may not need patents covering the underlying foundation model. It may instead build a portfolio around navigation, manipulation, safety, task planning or industrial workflows.
China has another advantage: deployment scale
China also has a large industrial robotics base.
The International Federation of Robotics reported more than 2 million industrial robots operating in Chinese factories in 2024, with about 295,000 new installations that year. China represented 54% of global industrial-robot installations in 2024.
In 2025, China’s share increased to 59% of global industrial-robot deployments.
This matters because physical AI depends on real-world deployment. Robots can generate information about manufacturing processes, objects, movement, maintenance, safety and human-machine interaction.
The potential cycle is:
AI models → physical deployment → operational data → improved AI → more capable machines → new inventions → new patents
This is one reason the intersection of GenAI and robotics could become an important IP battleground.
The next competition may be over the AI stack, not the model
Future disputes may not focus only on who owns a foundation model.
The more complicated question may be:
Who owns the technology required to turn AI intelligence into physical action?
An autonomous robot can involve sensors, sensor-fusion algorithms, multimodal AI, environment representation, task and motion planning, safety systems, real-time control, edge computing, communications and human-machine interaction.
Each layer can contain separate inventions.
This creates a dense patent environment. A company entering robotics, autonomous vehicles or industrial AI cannot rely on a search limited to “robotics.” Relevant patents may sit across AI, computer vision, control systems, telecommunications, sensors, semiconductors and industry-specific technologies.
That makes freedom-to-operate analysis increasingly important.
Patent volume will not equal patent strength
The size of the GenAI patent surge should not be treated as a direct measure of patent quality.
Patent counts do not establish claim strength, enforceability, commercial relevance or freedom-to-operate risk. There is also a major difference between domestic filing and international protection.
For companies assessing competitive positions, the more useful questions are:
- What is being claimed?
- Where is protection being sought?
- Which claims overlap with the technology?
- Which portfolios have commercially relevant protection?
- Where are the patent whitespace opportunities?
Trade secrets will remain part of the equation
Not every important AI innovation will appear in a patent.
Model weights, training datasets, evaluation methods, proprietary data pipelines and some model architectures may be difficult for competitors to detect. These can be candidates for trade-secret protection.
In physical AI, visible system-level innovations may be easier to reverse-engineer and therefore may favor patent protection, while hidden training processes or datasets may remain better suited to secrecy.
The strategic question is not simply how many patents to file.
It is which part of the technology should be patented, which should remain confidential, and where protection should be obtained.
Where the next IP battles could emerge
- AI-enabled robotics
Foundation models are moving toward robots that can understand natural-language instructions and perform more generalised tasks. The IP opportunity extends to perception, planning, manipulation and control.
- Autonomous vehicles
Generative models could influence planning, simulation, human-machine interaction and decision-making, creating overlap between existing autonomous-driving portfolios and newer GenAI patents.
- Industrial automation
Manufacturing is a major opportunity because AI can be integrated into production, inspection, maintenance and process optimisation.
- Multimodal and agentic systems
WIPO’s data shows rapid growth in multimodal technologies, while early patenting activity around agentic AI is emerging. As AI systems plan and execute sequences of actions, the boundary between GenAI and autonomous systems may become less clear.
- Edge and infrastructure AI
Physical systems cannot always rely on cloud processing. Latency, connectivity, power and safety requirements can push AI computation closer to the machine, creating IP opportunities around model compression, inference, hardware acceleration and real-time processing.
The IP map is changing
The GenAI patent surge is more than a story about China’s filing volume. It signals a change in where AI innovation is taking place.
The first phase of the GenAI patent race focused heavily on models, algorithms and content generation. The next phase is likely to spread across robots, vehicles, factories, infrastructure, sensors and autonomous machines.
China enters this phase with a large GenAI patent base, a rapidly expanding robotics ecosystem and the world’s largest industrial-robot deployment market. At the same time, patent activity is accelerating in the United States, Japan, Europe and other technology centres.
For technology companies, the question is no longer only who is building the best AI model.
It is increasingly who is protecting the technologies that allow AI to perceive, decide and act in the real world.
That is where the next generation of patent competition could emerge.
Sources
- https://www.wipo.int/web-publications/spark-patent-trends-update-in-genai/en/index.html
- https://www.wipo.int/web-publications/spark-patent-trends-update-in-genai/en/global-patenting-trends.html
- https://www.wipo.int/web-publications/spark-patent-trends-update-in-genai/en/executive-summary.html
- https://www.wipo.int/en/web/office-china/w/news/2026/wipo-china-genai-innovation-soaring-with-patent-activity-nearly-tripling-in-two-years
- https://www.wipo.int/pressroom/en/articles/2026/article_0012.html
- https://www.wipo.int/pressroom/en/articles/2024/article_0009.html
- https://www.wipo.int/en/web/patent-analytics/generative-ai/2025
- https://www.wipo.int/pressroom/en/articles/2026/article_0003.html
- https://www.epo.org/en/about-us/observatory-patents-and-technology/policy-and-funding/patents-and-standards/publications
- https://www.foley.com/insights/publications/2026/09/the-genai-patent-surge-is-coming-for-physical-ai-what-autonomous-systems-companies-should-be-watching/
- https://ifr.org/ifr-press-releases/news/global-robot-demand-in-factories-doubles-over-10-years
- https://ifr.org/news/five-million-robots-now-operate-in-factories-globally/
