AI & Machine Learning Patents: Top Companies & Filing Trends 2026
- 13.1% concentration at the very top. the five leading assignees together hold 76,282 of 582,780 records in scope, with a long tail of single- and few-filing entrants behind them.
- Filing volume is still climbing. filings rose from 2,455 in 2021 to 3,860 in 2024, a 57% increase over that span, with 2024 the most recent year that can be read as complete.
- Claim density clusters in one class. G06N (computing based on AI models) covers 2.0% of all records in scope, more than any other IPC subclass, while healthcare informatics and image-data generation sit well behind it.
Filing growth compares 2021 (2,455 records) with 2024 (3,860) — a three-year span. 2024 is the most recent year we treat as complete: publication lags filing by roughly 18 months, so 2025 onwards are still filling in and any growth rate that ends there would understate the field. Top-5 share is the combined record count of the five largest assignees divided by all 582,780 records in scope (CR5), not by the ranked leaders only.
What the AI and machine learning patent record shows
The dataset spans 582,780 published records filed between 2015 and mid-2026, drawn from a search string built around foundation models, generative AI, multimodal models, computer vision, reinforcement learning, autonomous agents and synthetic data. It captures both the core computing techniques behind AI systems and the application layers built on top of them, from image recognition to business-process automation. Publication lags filing by roughly 18 months, so the final one or two years in any trend line will always look lighter than they eventually turn out to be.
Reading this landscape means separating three questions that are easy to conflate: who files the most, who files the fastest right now, and where claim space is still thin. The ranking answers the first, recent-year momentum the second, and the IPC composition and receiving-office breakdown the third. None of the three alone tells a full story about where to file next.
Filing trend and technology composition
Two views of the same 582,780-record dataset: filing volume by year, and the IPC subclasses that carry the claim density. Because a single record can carry several classes, the composition shares add up to more than the record total.
Filing trend, 2017-2026
Annual filings grew from 323 in 2017 to a peak of 3,860 in 2024. The 2021-to-2024 span alone saw a 57% increase. 2025 and 2026 figures (287 in the latest partial year) understate real activity because of publication lag, not a slowdown in filing.
IPC subclass composition
G06N (computing arrangements based on AI models) is the largest single subclass at 2.0% of all records, ahead of G06F general digital data processing at 1.2%. Application-layer classes — image/video recognition, business-process computing, image data generation, healthcare informatics — each sit at 0.3-0.4% of records, indicating that core computing claims are more heavily contested than most downstream application claims.
Shares are the percentage of the 582,780 records in scope. A patent can carry several IPC classes, so the shares add up to more than 100%.
Go deeper on Artificial Intelligence & Machine Learning Patent Landscape with Eureka
This page is one run against one query. Ask Eureka your own question about artificial intelligence & machine learning patent landscape and every answer comes back with the patent numbers behind it.
Try EurekaRepresentative filing and most-cited records
Artificial intelligence for synthetic data generation (US20260236235A1)
The technique generates a natural-language workflow prompt instructing a generative machine learning model to produce multiple versions of workflow code for synthetic data generation. At least one evaluator machine learning model then selects among those versions, and the selected workflow code is executed to generate the synthetic data.Filed by International Business Machines Corporation; the claim structure pairs a generator model with a separate evaluator model in a workflow-selection loop.
| # | Publication no. | Patent title | Citations |
|---|---|---|---|
| 1 | US20150379430A1 | Efficient duplicate detection for machine learning data sets | 857 |
| 2 | US20220126864A1 | Autonomous vehicle system | 680 |
| 3 | US20150379429A1 | Interactive interfaces for machine learning model evaluations | 596 |
| 4 | US20170124487A1 | Systems, methods, and apparatuses for implementing machine learning model training and deployment with a roll… | 399 |
| 5 | US20190236598A1 | Systems, methods, and apparatuses for implementing machine learning models for smart contracts using distribu… | 398 |
| 6 | US20240386015A1 | Composite symbolic and non-symbolic artificial intelligence system for advanced reasoning and semantic search | 378 |
| 7 | US20160358099A1 | Advanced analytical infrastructure for machine learning | 360 |
| 8 | US20190384303A1 | Behavior-guided path planning in autonomous machine applications | 356 |
| 9 | US20220187847A1 | Robot Fleet Management for Value Chain Networks | 341 |
| 10 | US20210133670A1 | Control tower and enterprise management platform with a machine learning/artificial intelligence managing sen… | 340 |
Citation counts favour older records simply because they have had longer to accumulate them inside this corpus; treat them as a signal of influence within the searched set, not as a ranking of present-day importance.
Each row carries its publication number; clicking a row searches Eureka by that number.
Put your own technology through the same analysis
Eureka on the web
When you want the answer in the next five minutes.
The agent works the prompt against patents and technical literature, citing every source.
Run your analysis now →MCP server & REST API
When it has to run inside your own pipeline.
Patent search, landscape analysis and assignee resolution as MCP tools. Drop them into any agent framework, or call REST directly.
Browse MCP servers →What the concentration and momentum data mean
The ranking, the growth curve and the recent-year momentum figures each answer a different question. Read together, they point to a field where the leading position is contested but not locked down, and where several established filers have pulled back sharply in the most recent year.
The top of the field is crowded but not dominant
The five leading assignees together account for 76,282 records, 13.1% of everything in scope; the top ten add up to 19.3%. That leaves roughly four-fifths of the record base spread across a long tail of filers, which is unusual for a field this large and suggests the leadership position is still winnable rather than settled.
Filing momentum is real, not an artefact of publication lag
Annual filings climbed from 2,455 in 2021 to a peak of 3,860 in 2024, the most recent year that can be treated as complete. That three-year increase came before the current wave of generative-AI product launches fully worked through the roughly 18-month gap between filing and publication, so the underlying filing rate for 2025 is likely understated in the raw counts shown today.
Several heavy filers pulled back sharply in the latest year
Assignees that filed heavily in prior years show steep year-on-year declines in the latest tracked year — drops in the 59% to 93% range across the companies with the largest recent-year pullbacks. This looks more like a publication-lag effect on very recent filings than a genuine retreat from the technology, but it means recent-year rankings should not be read as a real-time leaderboard.
The United States dominates the receiving-office mix
United States filings (15,675) outnumber the next-largest office, WIPO/PCT (2,028), by nearly eightfold, with Europe, Australia, Canada and the United Kingdom each accounting for smaller slices. Anyone building a freedom-to-operate view for this field needs to start with US prosecution before weighing PCT or EPO strategy.
Eureka can read the same corpus for gaps instead of for coverage: under-claimed branches adjacent to artificial intelligence & machine learning patent landscape, with the prior art for and against each one.
Who is filing, and where the gate sits
The 100-company ranking spans large diversified technology firms filing across many application areas alongside a long tail of narrower filers. Co-assignee pairs are rare in this dataset — only 10 identified — which suggests most filings here come from internal R&D rather than joint development.
A single leader holds a clear but not overwhelming lead
The top-ranked assignee's 27,644 records sit well ahead of fifth place at 10,458, but the gap to the rest of the ranked field narrows quickly after that, and the top 5 combined still represent only 13.1% of all records in scope.
Most of the field sits outside the ranked leaders
The ten most active assignees combine for 19.3% of the 582,780 records in scope, meaning roughly four out of five records come from assignees outside that group — a wide base of corporate, academic and individual filers rather than a small closed set of dominant players.
Joint filing is the exception, not the norm
Only 10 co-assignee pairs appear across the dataset, with the strongest pairing linked to affiliated entities of the same corporate group rather than independent partners. That points to a field built mainly on solo corporate R&D programmes rather than cross-company alliances.
| Assignee | Recent year | YoY |
|---|---|---|
| NVIDIA Corporation | 10 | -91% |
| Capital One Services, LLC | 7 | -59% |
| Qualcomm Incorporated | 6 | -93% |
| Oracle International Corporation | 6 | -81% |
| Bank of America Corporation | 5 | -81% |
| Microsoft Technology Licensing, LLC | 3 | -93% |
| Google LLC | 3 | -83% |
| SAP SE | 2 | -82% |
Where to take this analysis
The figures here describe the shape of the field as it stands in the current data cut. Turning that into a filing or freedom-to-operate decision means going deeper on specific claim sets and specific competitors.
Check freedom-to-operate against the leading assignees
Concentration at the top is modest, but the leading assignees still hold dense claim clusters in G06N and G06F. A targeted search against their specific claim language is worth running before committing to a filing strategy in adjacent territory.
Run a freedom-to-operate search in EurekaTrack momentum, not just cumulative rank
Several previously heavy filers show sharp year-on-year pullbacks in the latest tracked year, largely a publication-lag artefact. Watching quarterly filing behaviour rather than annual totals gives an earlier read on where competitive filing pressure is actually building.
Set up a monitoring alert in EurekaMap the under-claimed branches before they fill in
Evaluator-model selection loops and synthetic-data workflow generation show comparatively thin claim density today. That is a narrow window, and dense core classes like G06N suggest the broader field is being claimed quickly.
Explore white space in EurekaCommon questions about AI and machine learning patents
The dataset's assignee ranking is led by a single company with 27,644 records, well ahead of the fifth-ranked assignee at 10,458. That said, the top five assignees combined hold only 13.1% of the 582,780 records in scope, and the top ten hold 19.3%, so no single company or small group controls a majority of the filing activity. The remaining share is spread across a long tail of corporate, academic and individual filers ranked further down the list.
Yes, based on complete years of data. Annual filings rose from 2,455 in 2021 to 3,860 in 2024, a 57% increase, and 2024 is the peak year and the most recent year that can be read as complete. Figures for 2025 and 2026 look lower in the raw data, but that reflects the roughly 18-month lag between filing and publication rather than an actual slowdown — those years will fill in as more records publish.
G06N, covering computing arrangements based on specific computational models like neural networks, is the largest single IPC subclass in this dataset at 2.0% of all 582,780 records. G06F, general electric digital data processing, follows at 1.2%. Application-specific classes such as image and video recognition (G06V), business-process computing (G06Q), image data processing (G06T) and healthcare informatics (G16H) each sit at 0.3% to 0.4% of records, meaning core computational claims are more densely contested than most downstream application areas.
The receiving-office data in this dataset shows the United States as by far the largest single office, with 15,675 filings, compared with 2,028 at WIPO under the PCT and 1,589 at the EPO. Australia, Canada and the United Kingdom each account for a much smaller share. Any freedom-to-operate or filing-strategy review for this field should therefore start with US prosecution history before extending to PCT or European coverage.
The clearest signal is relative: core computing claims in G06N and G06F are far denser than application-layer classes like healthcare informatics (G16H) or image data generation (G06T), which sit at roughly a fifth of the density of the core classes. Within recent filings, specific technical mechanisms such as evaluator-model selection loops paired with generator models, and synthetic-data workflow generation, appear as narrower, more recently opened claim territory. High density in the core classes signals that space is occupied, not that it is technically exhausted, so a narrow, well-drafted claim can still find room even in a crowded subclass.
Research Artificial Intelligence & Machine Learning Patent Landscape in depth with Eureka
Go past this page: query the whole artificial intelligence & machine learning patent landscape corpus yourself, in your own scope.
Every answer comes back with patent numbers you can open.
Disclaimer. This page is generated from Patsnap Eureka data drawn from a limited snapshot of global patent and scientific-literature records, and is provided for general information and reference only.
Patent data carries inherent limitations: recent filings (typically the most recent 18–24 months) are under-counted due to standard publication lag; counts may be reported at either a patent-family or a patent-record basis and are not always directly comparable; classification, applicant-name, and citation data may contain errors, duplicates, or omissions; and the underlying search query defines and constrains the scope shown. As a result, the analysis may be incomplete or inaccurate and may not reflect the full technology landscape.
Nothing on this page constitutes an exhaustive prior-art, novelty, freedom-to-operate, or validity search, nor does it constitute legal, financial, investment, or professional advice, and it should not be relied upon as such. Any patent, commercial, or strategic decision should be verified independently and reviewed with qualified patent, legal, and domain professionals. Patsnap makes no warranties, express or implied, as to the accuracy, completeness, or fitness for any particular purpose of the information presented.
Machine translation. Assignee and organisation names originally recorded in Chinese, Japanese or Korean have been rendered into English by an AI translation step so that the tables stay readable. These renderings are best-effort and may not match a company’s registered English name; the original name is what the underlying patent record carries, and it is what any Eureka query launched from this page uses.