Table of Contents
Few areas of technology have reshaped the patent map as quickly as machine learning, and the AI patent filing trends emerging from WIPO and the USPTO make the scale hard to overstate: annual generative-AI patent families climbed from fewer than 800 in 2014 to more than 14,000 in 2023, and the pace has only accelerated since. This guide pulls the primary numbers together, names the organizations driving them, and explains what the trajectory means if you are deciding where and what to file.
AI Patent Filing Trends at a Glance
The headline AI patent filing trends come from two authoritative sources: the World Intellectual Property Organization’s Patent Landscape Report on Generative AI and the USPTO’s Inventing AI benchmark study. Read together, they tell a consistent story: AI is not a niche filing category any more — it is diffusing across the whole patent system.
- 54,358 generative-AI patent families were published between 2014 and 2023 (WIPO), 89% of them still active at the end of 2023.
- Annual GenAI families rose from under 800 in 2014 to more than 14,000 in 2023 — roughly 45% average annual growth since 2017.
- More than 56,000 new GenAI families were published in 2024 and 2025 alone, exceeding the entire prior decade (WIPO SPARK update).
- At the USPTO, AI-related applications more than doubled from about 30,000 (2002) to over 60,000 (2018), and AI’s share of all applications grew from 9% to nearly 16%.
- AI now appears in more than 42% of all technology subclasses, up from 9% in 1976.
The GenAI Surge: From 800 to 14,000 Families a Year
The single most striking of these trends is the shape of the curve. WIPO identified 54,358 generative-AI patent families published across the ten years to 2023, but that total is front-loaded toward the end of the period. Fewer than 800 families were published in 2014; by 2023 the annual figure had passed 14,000. The inflection came around 2017 — the year the transformer architecture that underpins modern large language models was introduced — after which average annual growth ran near 45%.
That acceleration has not levelled off. WIPO’s later SPARK trends update recorded more than 56,000 new generative-AI patent families in 2024 and 2025 combined — more, in two years, than the field produced in the whole decade the original report covered. For anyone building an IP position in this space, the practical consequence is that a landscape snapshot ages in months, not years, and any filing strategy has to assume the surrounding art is still expanding rapidly.
Who Is Filing: The Assignees Behind the Numbers
AI patenting is dominated by a small group of very large organizations. In WIPO’s ranking of generative-AI applicants, the leaders are Tencent, Ping An Insurance Group and Baidu, with IBM, Samsung Electronics, Alphabet (Google) and Microsoft all inside the global top ten. These are parties with both the portfolios and the willingness to enforce them.
A second tier is easy to misread: research institutions such as the Chinese Academy of Sciences (more than 600 families), Tsinghua University and Zhejiang University file heavily but rarely assert their patents. When you analyse these filings for competitive purposes, separating operating companies that sue from universities that publish is essential — a raw applicant ranking overstates the litigation risk from the academic filers and understates the strategic weight of the corporate ones.
- Corporate leaders: Tencent, Ping An, Baidu, IBM, Samsung, Alphabet, Microsoft.
- Heavy academic filers: Chinese Academy of Sciences, Tsinghua, Zhejiang University.
- Takeaway: rank by ultimate parent, then split assertive filers from non-asserting ones before drawing conclusions.
Where the Filings Originate: China’s ~70% Share
Geography is where this analysis most often trips up a casual reader. WIPO’s data attributes roughly 38,000 generative-AI patent families to inventors based in China — about 70% of the global total — growing near 50% a year. The United States is a distant second at around 6,300 families, followed by the Republic of Korea, Japan and India, the last of which is growing fastest of all at about 56% annually.
By office of first filing, China’s national office received more than 40,000 of the families, the USPTO more than 10,700, the PCT route more than 7,700, and the European Patent Office more than 3,100. The lesson for anyone reading these trends is blunt: a landscape built only on English-language U.S. patents captures a minority of the field and can label already-crowded technical areas as open white space.
AI’s Diffusion Across Technology (USPTO Data)
The USPTO’s Inventing AI study measured a different but complementary dimension of AI patenting: not just how many AI applications there are, but how widely AI has spread. Between 2002 and 2018, annual AI-related applications more than doubled, from roughly 30,000 to over 60,000, and AI’s share of all applications rose from 9% to nearly 16%.
More telling is the diffusion figure. AI appeared in about 9% of technology subclasses in 1976; by 2018 it had spread to more than 42% of them. The USPTO’s Artificial Intelligence Patent Dataset now classifies AI across 15.4 million U.S. patent documents using eight AI component technologies — a reminder that AI is increasingly a feature of inventions in other fields (automotive, medical devices, telecoms) rather than a standalone category. Whatever industry you file in, AI-related prior art is now part of your landscape.
Reading the Data With Care
A few caveats keep these numbers honest. Patent families are counted by first publication, and applications generally publish about 18 months after filing, so the most recent years are always understated until pending cases surface — the 2025 and 2026 totals will climb further as those applications appear. Counts also differ by database and by how each source defines “AI”: WIPO’s generative-AI scope is deliberately narrower than the USPTO’s eight-component definition, so the two figures measure different universes and should never be added together.
Filing volume is also not the whole innovation picture. Some AI developers deliberately keep core model architectures and training techniques as trade secrets rather than disclosing them in a published patent, which means the record understates real activity in the fastest-moving areas. Treat these figures as a well-sourced floor and a directional signal about where innovation and enforcement risk are concentrating — not as a precise census of everything being built.
How Many AI Patents Are Filed Each Year?
Because the question comes up constantly, it is worth answering directly. There is no single global counter, but the data give tight ranges. For generative AI specifically, WIPO’s data shows more than 14,000 patent families published in 2023, and more than 56,000 new families across 2024 and 2025. For AI broadly at the USPTO, applications passed 60,000 per year as of 2018 and have continued climbing. The numbers differ because they measure different things — generative-AI families worldwide versus all AI-related applications at one office — so always check the scope before comparing two figures.
What the Trends Mean for Your Filing Strategy
Read strategically, the AI patent filing trends point to four practical moves. First, file where the art is genuinely thin: model-training methods are densely patented by the incumbents, while adjacent areas — retrieval orchestration, evaluation and guardrail tooling, data provenance — remain comparatively open. Second, search by classification, not keywords, because GenAI terminology shifts monthly. Third, include Chinese-origin filings, or your landscape is a blind spot dressed as good news. Fourth, monitor continuously, since 56,000 new families in two years means anything you found last quarter is already incomplete.
This is exactly the sequence we walk through in our AI patent landscape analysis case study, where a competitive-intelligence review mapped 1,240 competitor families and located five white-space clusters before a fundraise. If you want that applied to your own roadmap, our competitive IP intelligence and AI market-research services are built for it.
Turn AI Patent Filing Trends Into a Filing Plan
PerspireIP turns the raw data into a defensible landscape: who holds what, where the white space is, and where you are exposed. Talk to our market-research team about a landscape scoped to your technology.
Frequently Asked Questions
How many AI patents are filed each year?
There is no single global figure, but WIPO recorded more than 14,000 generative-AI patent families published in 2023 and over 56,000 new families across 2024-2025. At the USPTO, AI-related applications passed 60,000 per year as of 2018 and have kept rising.
Which companies file the most AI patents?
WIPO’s generative-AI ranking is led by Tencent, Ping An Insurance Group and Baidu, with IBM, Samsung, Alphabet (Google) and Microsoft in the global top ten. Research institutions such as the Chinese Academy of Sciences also file heavily but rarely assert their patents.
Which country files the most AI patents?
China dominates, with inventors based there accounting for roughly 70% of generative-AI patent families — about 38,000 — versus around 6,300 for the United States. Any credible AI landscape has to include Chinese-origin filings.
Why are AI patent filings growing so fast?
Generative-AI patenting inflected around 2017, when the transformer architecture appeared, and has grown near 45% a year since. The surge reflects both genuine research investment and a race among large incumbents to build defensive and offensive portfolios.
How do AI patent filing trends affect my strategy?
They tell you to file into thin, adjacent white space rather than crowded training-method art, to search by patent classification rather than shifting keywords, to include non-English filings, and to monitor continuously because the landscape changes every quarter.