Large Language Models Patent Landscape 2026
Large Language Models Patent Landscape in 2026
The large language models patent field spans 19,964 patent families and is in a clear growth stage, with filings accelerating sharply since 2021 and the most recent period still understated by publication lag. Microsoft Technology Licensing leads by family count, but the top five filers account for only about a quarter of the hundred largest filers’ combined output, pointing to a competitive, multi-player field rather than a true monopoly.
Microsoft leads a crowded field of US, Chinese, and European applicants
Microsoft Technology Licensing LLC heads the top tier is genuinely competitive: the gap between first and second place is narrow relative to the corpus size, and
The top five filers together hold 26% of the hundred largest filers’ combined total — a moderate concentration level. No single player commands a dominant share, and Chinese applicants (Tencent, Baidu, Alipay, Huawei, China Mobile) occupy four of the top ten slots alongside US and European incumbents, indicating a genuinely global contest.
| # | Applicant | Patent families | Share |
|---|---|---|---|
| 1 | Microsoft Technology Licensing LLC | 1,320 | |
| 2 | Google LLC | 1,186 | |
| 3 | Tencent Technology (Shenzhen) Co., Ltd. | 800 | |
| 4 | BEIJING BAIDU NETCOM SCI & TECH CO LTD | 677 | |
| 5 | Alipay (Hangzhou) Information Technology Co., Ltd. | 538 | |
| 6 | Zhejiang University | 433 | |
| 7 | Qualcomm Incorporated | 431 | |
| 8 | Huawei Technologies Co., Ltd. | 399 | |
| 9 | Samsung Electronics Co., Ltd. | 358 | |
| 10 | China Mobile Communications Group Co., Ltd. | 349 |
| # | Applicant | Patent families | Share |
|---|---|---|---|
| 11 | International Business Machines Corporation | 342 | |
| 12 | NVIDIA Corporation | 342 | |
| 13 | Tsinghua University | 330 | |
| 14 | Intuit Inc. | 328 | |
| 15 | GDM Holding LLC | 320 | |
| 16 | Salesforce, Inc. | 293 | |
| 17 | iFLYTEK Co., Ltd. | 277 | |
| 18 | China Telecom Corporation Limited | 219 | |
| 19 | Alibaba (China) Co., Ltd. | 218 | |
| 20 | Robert Bosch GmbH | 209 |
The breadth of the ranking — spanning hyperscalers (Microsoft, Google), semiconductor designers (Qualcomm, NVIDIA), telecoms (China Mobile, China Telecom), financial-technology platforms (Alipay, Intuit), and research universities (Zhejiang, Tsinghua) — signals that LLM IP is being accumulated across the full stack, from model architecture to vertical application.
Filings from the most recent 18–24 months are under-represented due to patent-publication lag; current-year and prior-year totals should be treated as lower bounds. Longer-window growth, applicant concentration, and technology-route coverage are therefore more reliable signals than the latest-year bar alone.
Explosive filing growth since 2021, anchored in digital data processing and AI model classes
Annual filing volume and the IPC class breakdown together reveal both the pace of investment and the technical areas receiving the most attention; the trend chart in particular shows a step-change in activity beginning in 2022–2023.
Annual filing trend
Filings were negligible before 2020 (2 families in 2017), then accelerated sharply: 393 in 2021, 771 in 2022, 3,657 in 2023, and 8,559 in 2024. The 2025 figure of 6,083 and the 2026 figure of 295 are understated due to publication lag and should not be read as a slowdown. The overall recent-window growth rate is 2,135 patent families, confirming the field remains in a growth stage.
↗ Hover for values · click a bar to ask EurekaTechnology composition
G06F (electric digital data processing) and G06N (AI model computing) together account for the large majority of classified records, reflecting the model-architecture and inference-optimization core of LLM IP. Adjacent classes — G06Q (business/commerce), H04L (digital transmission), G06V (image/video recognition), G10L (speech/audio), and G16H (healthcare informatics) — each represent meaningful but smaller slices, indicating active application-layer development across verticals.
↗ Hover for values · click a bar to ask EurekaHighly cited patent families surfaced by the query
Citation-heavy patent families returned by the query. Use this section as citation context, not as a curated list of the most topic-specific patents.
Language capability evaluation of large language m…
Systems and methods for language capability evaluation of large language models are provided. A communication platform accesses a pair of parallel inputs, including a reference input in a reference language corresponding to a target input in a target language different from the reference language. The communication platform executes a reference large… (excerpt from the patent abstract)
| # | Patent | Citations |
|---|---|---|
| 1 | Robot Fleet Management for Value Chain Networks | 322 |
| 2 | Composite symbolic and non-symbolic artificial int… | 321 |
| 3 | Systems and methods for machine content generation | 256 |
| 4 | System and method for improving text input in a sh… | 235 |
| 5 | Transaction platforms where systems include sets o… | 217 |
| 6 | Image composites using a generative adversarial ne… | 211 |
| 7 | Ultra large language models as ai agent controller… | 210 |
| 8 | Cross-context natural language model generation | 207 |
Ranked by total forward citations. Citation counts favour older and broadly cited patent families, and broad or adjacent patents may appear when they match the search scope. Treat this section as citation context, not as a curated list of the most topic-specific patents.
What the patent structure means for R&D strategy
The combination of rapid growth, moderate concentration, cross-sector collaboration, and US-dominant jurisdiction coverage creates both competitive pressure and identifiable strategic options for new entrants and incumbents alike.
Growth stage: filings still accelerating
The field is classified as Growth, with annual filing volume rising steeply across the multi-year window and no evidence of a peak or plateau. The 2024 cohort of 8,559 families is the largest single-year total on record. Publication lag means the 2025 and 2026 numbers will revise upward materially, so the apparent recent softening is an artifact of data timing rather than a real deceleration. R&D investment entering now will still encounter an open, fast-moving competitive environment rather than a saturated one.
Growth stageMulti-tier field: strong leaders but no monopolist
The top five filers represent 26% of the hundred largest filers’ combined output, a moderate figure that leaves roughly three-quarters distributed across a large secondary tier. Microsoft and Google lead, but the gap to Tencent, Baidu, and Alipay is not prohibitive. A well-resourced challenger entering a specific vertical or architecture niche can build a defensible position without confronting a single locked-in incumbent. The presence of university applicants — Zhejiang, Tsinghua, Peking, Shanghai Jiao Tong — also signals that academic-origin IP remains a relevant licensing or acquisition vector.
Moderate concentrationIBM and Samsung drive the most active co-filing partnerships
The most active co-applicant pair is DeepMind Technology and GDM Holding LLC, with 45 jointly filed families. IBM (International Business Machines) is the next most prolific co-filer, partnering with IBM United Kingdom (10 families), IBM Deutschland (8), MIT (5), IBM Israel (5), and IBM China (3). Samsung Electronics co-files with Naver Corporation (3 families), KAIST (2), and Hanyang University (2). These patterns suggest that hardware-software integration and academic-industry translation are the primary drivers of formal collaboration in this space.
Co-filing activeUS is the primary filing office; PCT and China are secondary channels
The United States accounts for the largest share of jurisdictional records, followed by WIPO PCT, China, and Europe (EPO). Canada, Australia, and India each show meaningful but smaller volumes, reflecting secondary validation markets. The heavy US weighting means freedom-to-operate analysis must prioritize USPTO prosecution, but the substantial PCT volume indicates applicants are actively preserving global optionality. Japan and South Korea, despite hosting major technology companies, show comparatively low direct filing counts in this corpus.
US-led, global reachGo beyond the landscape: Eureka’s TRIZ Solution agent breaks down an R&D problem and returns patented concept solutions, each with a technical approach and cited patent & literature evidence.
| Applicant | Collaborator | Co-filings |
|---|---|---|
| DeepMind Technologies Limited | GDM Holding LLC | 45 |
| International Business Machines Corporation | IBM UNITED KINGDOM LTD INTELLECTUAL PROPERTY DEPAR… | 10 |
| International Business Machines Corporation | IBM DEUTSCHLAND GMBH | 8 |
| International Business Machines Corporation | Massachusetts Institute of Technology | 5 |
| International Business Machines Corporation | IBM ISRAEL SCI & TECH LTD | 5 |
| International Business Machines Corporation | IBM China Co., Ltd. | 3 |
| Samsung Electronics Co., Ltd. | Naver Corporation | 3 |
| International Business Machines Corporation | École Polytechnique Fédérale de Lausanne (EPFL) | 2 |
| Samsung Electronics Co., Ltd. | Korea Advanced Institute of Science and Technology (KAIST) | 2 |
| Samsung Electronics Co., Ltd. | Hanyang University | 2 |
Co-filing pairs, ranked by the number of jointly-filed patent families.
Microsoft and Google lead on volume; Qualcomm and NVIDIA emerge as fast-rising challengers
The two largest filers share broadly similar IPC profiles — both weighted toward natural language processing (G06F 40) and neural-network architectures (G06N 3) — but Google’s recent momentum is markedly higher, and hardware-focused challengers are entering the ranking from a near-zero base.
Microsoft Technology Licensing LLC
Microsoft Technology Licensing LLC holds 1,320 patent families, the largest portfolio in the corpus. Its technology emphasis is concentrated in natural language processing (G06F 40, 774 families) and neural-network computing (G06N 3, 476 families), with a secondary cluster in data retrieval and management (G06F 16, 319 families). Recent-period momentum is strong at 9.6× versus the prior three-year window, reflecting broad investment across LLM training, fine-tuning, and application integration.
families: 1,320Google LLC
Google LLC ranks second with 1,186 patent families and shows the highest momentum among established leaders at 34× versus its prior three-year filing rate — a step-change that reflects the rapid deployment of transformer-based products. Google’s emphasis skews slightly more toward AI model architecture (G06N 3, 542 families) than toward NLP processing (G06F 40, 500 families), suggesting deeper investment in foundational model research relative to Microsoft’s application-layer weighting. The 34× figure originates from a comparatively small prior base, but the absolute recent volume of 805 families confirms genuine scale.
families: 1,186| Applicant | Recent (3 yrs) | Trend |
|---|---|---|
| Microsoft Technology Licensing LLC | 948 | ▲ 9.6× vs prior 3-yr |
| Google LLC | 805 | ▲ 34× vs prior 3-yr |
| International Business Machines Corporation | 272 | ▲ 8.0× vs prior 3-yr |
| Intuit Inc. | 226 | ▲ new entrant |
| Qualcomm Incorporated | 214 | ▲ new entrant |
| DeepMind Technologies Limited | 235 | ▲ 11× vs prior 3-yr |
| NVIDIA Corporation | 212 | ▲ new entrant |
| GDM Holding LLC | 185 | ▲ 15× vs prior 3-yr |
Under-served adjacent branches worth monitoring
Five IPC classes sit adjacent to the dominant G06F/G06N core with comparatively lower filing shares; each has plausible technical synergies with LLMs and may represent areas where early IP positioning remains feasible.
G10L · Speech and audio analysis/synthesis
G10L holds a 3% share of classified records in this corpus — notably sparse given the direct relevance of speech recognition, text-to-speech synthesis, and voice-based LLM interfaces to deployed products. The technical entry path is well-defined: combining transformer-based audio encoders with LLM decoders for end-to-end spoken dialogue systems. Applicants with existing acoustic-model portfolios (e.g., in telephony or automotive voice) could leverage them to differentiate in a relatively less contested IPC sub-space.
Search this in Eureka →G16H · Healthcare informatics
G16H accounts for a modest share of records despite growing clinical interest in LLMs for clinical note summarization, diagnostic support, and patient-communication automation. The branch is sparse relative to the overall corpus size — under 900 records in a 19,964-family dataset — suggesting that healthcare-specific LLM IP has not yet been filed at scale, possibly reflecting regulatory caution. Applicants with existing medical-device or health-IT portfolios and access to clinical validation data are best positioned to build defensible claims here before the space fills.
Search this in Eureka →How leading applicants differ by technology route
Strength of each leader across the main technology routes.
| Player | G06F 40 · Electric digital data processing | G06N 3 · Computing based on AI models | G06N 20 · Computing based on AI models | G06F 16 · Electric digital data processing | G06N 5 · Computing based on AI models |
|---|---|---|---|---|---|
| Microsoft Technology Licensing LLC | Strong · 774 | Strong · 476 | Moderate · 266 | Moderate · 319 | Emerging · 46 |
| Google LLC | Strong · 500 | Strong · 542 | Moderate · 187 | Moderate · 214 | Emerging · 24 |
| Intuit Inc. | Strong · 209 | Strong · 123 | Moderate · 82 | Moderate · 89 | Emerging · 30 |
| International Business Machines Corporation | Strong · 167 | Strong · 141 | Moderate · 56 | Moderate · 55 | Absent |
| GDM Holding LLC | Strong · 167 | Strong · 122 | Absent | Strong · 85 | Absent |
| Qualcomm Incorporated | Absent | Strong · 244 | Moderate · 62 | Absent | Absent |
| Samsung Electronics Co., Ltd. | Strong · 119 | Strong · 181 | Absent | Absent | Absent |
Frequently asked questions
The corpus contains 19,964 patent families in scope. Annual filing volume was negligible before 2020 and reached 8,559 families in 2024 alone, indicating the field is in a rapid-growth stage.
Microsoft Technology Licensing LLC leads with 1,320 patent families, followed by Google LLC (1,186), Tencent Technology Shenzhen (800), Beijing Baidu Netcom Science and Technology (677), and Alipay Hangzhou Information Technology (538). The top five together represent 26% of the hundred largest filers’ combined output.
The United States is the primary filing office, followed by WIPO PCT, China, and Europe (EPO). Canada, Australia, and India each register meaningful secondary volumes. Japan and South Korea show comparatively lower counts in this corpus despite hosting major technology companies.
G06F (electric digital data processing) and G06N (computing based on AI models) dominate the classified records. Adjacent classes with meaningful filing counts include G06Q (business and commerce data processing), H04L (digital information transmission), G06V (image and video recognition), G06T (image data processing), and G10L (speech and audio analysis).
Among established leaders, Google LLC shows the highest recent-period momentum at 34× versus its prior three-year filing rate, followed by an entity identified as GDM Holding LLC at 15×, DeepMind Technology at 11×, and Microsoft Technology Licensing at 9.6×. Qualcomm, Intuit, and NVIDIA are classified as new entrants in the recent period.
G10L (speech and audio analysis/synthesis) and G16H (healthcare informatics) are the most notable under-served adjacent branches relative to the overall corpus size. Both have clear technical synergies with LLMs but lower filing densities, suggesting that early IP positioning may still be feasible for focused entrants.
Ready to map your own LLM patent landscape?
Join 18,000+ innovators using PatSnap Eureka to map any technology landscape: search 2B+ patents and papers, surface key assignees, and generate a report like this in minutes.
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.