Global Generative AI Patents in 8 Numbers: China’s 6-to-1 Lead and the Surge to 37,000 Inventions

The global race for artificial intelligence dominance is frequently measured by venture capital rounds, compute cluster sizes, and foundation model leaderboards. Yet beneath the consumer interface of chatbots and image generators, an intense legal enclosure of intellectual property is underway. According to the United Nations World Intellectual Property Organization (WIPO) in its comprehensive study China-Based Inventors Filing Most GenAI Patents, WIPO Data Shows and its flagship report Patent Landscape Report – Generative Artificial Intelligence (GenAI), generative AI patents have exploded from an obscure technical niche into the fastest-growing segment of global computing technology.

Between 2014 and 2023, innovators filed 54,000 generative AI inventions (measured as simple patent families to eliminate cross-jurisdiction duplicates) and published more than 75,000 scientific research papers. Rather than tapering off as foundation models mature, the pace of intellectual property claims has intensified. In its subsequent analysis The GenAI Landscape Is Getting Bigger And More Competitive According To New WIPO Report, WIPO established that published GenAI patent families more than doubled between 2023 and 2025, surging past 37,000 annual publications. Here are the eight numbers that explain who is patenting generative artificial intelligence, what technologies they are locking down, and how the global intellectual property balance is shifting.

1. 54,000 inventions: The ten-year baseline that set the foundation

In its baseline landscape analysis, WIPO documented 54,000 GenAI inventions filed across the globe between 2014 and 2023. Over that decade, patenting activity evolved from experimental computer vision prototypes into commercial machine learning systems.

Patenting activity remained modest during the early years of deep learning. In 2014, global inventors published only 733 GenAI patent families. Following the 2017 introduction of the deep neural network transformer architecture—which provided the structural breakthrough for modern large language models—filings accelerated dramatically. WIPO documents that generative AI patenting expanded eightfold after 2017.

By 2023, annual publications reached 14,080 patent families. In that single year, over 25% of all GenAI patents and over 45% of all scientific research papers accumulated over the entire decade were published.

This patent boom accompanied a massive surge in private research capital, as documented in our tracking of Stanford AI Index 2026 benchmarks. While algorithmic advances were initially shared through open pre-prints and open-source repositories, commercialization prompted tech giants and enterprise incumbents to rapidly construct defensive patent thickets.

2. 38,210 inventions: China’s overwhelming geographic lead

The geographic distribution of generative AI patents overturns common assumptions about American technological hegemony. Between 2014 and 2023, inventors based in China published 38,210 GenAI patent families—accounting for almost 70% of all generative AI inventions worldwide.

The United States ranked second with 6,276 patent families, representing an 11% world share. Our calculation: Chinese inventors published 6.1 times more generative AI patent families than their American counterparts over the ten-year period.

Source: WIPO, China-Based Inventors Filing Most GenAI Patents, WIPO Data Shows.

As WIPO details in Chapter 2 of its report, Global patenting and research in GenAI, China has published more generative AI patents each year since 2017 than all other countries combined. Furthermore, China achieved an average annual growth rate of 50% in patent publications between 2014 and 2023, despite starting from a substantially larger portfolio base.

By contrast, US patent publications grew at an average annual pace of 24% over the same period. While American firms lead the market in multi-billion-dollar capital expenditures across cloud infrastructure, Chinese enterprises and state-backed research institutions have prioritized patent volume across domestic industrial applications, consumer software, and telecom infrastructure.

3. 2,074 patents: Tencent and corporate giants lead the corporate pack

Enterprise dominance in generative AI patenting is concentrated among Asian technology conglomerates. The top corporate patent applicant worldwide is Chinese social media and gaming giant Tencent, which published 2,074 GenAI patent families between 2014 and 2023.

Tencent has deployed these patented architectures across its Hunyuan foundation model family, integrating text generation, image creation, and automated computer vision directly into WeChat and its interactive entertainment platforms.

The global corporate leaderboard highlights the industrial diversity of generative AI:

• Ping An Insurance Group ranks second with 1,564 patent families, applying algorithmic models to automated underwriting, risk assessment, and claims verification.
• Search engine pioneer Baidu ranks third with 1,234 patent families, powering its ERNIE foundation models across enterprise search, energy logistics, and autonomous transportation.
• The Chinese Academy of Sciences ranks fourth with 607 patent families, standing as the only university or research organization inside the global top ten.
• US computing pioneer IBM ranks fifth with 601 patent families, focused on enterprise watsonx governance and compliant multi-cloud workflows.
• E-commerce titan Alibaba Group ranks sixth with 571 patent families.
• Consumer electronics giant Samsung Electronics ranks seventh with 468 patent families.
• Alphabet ranks eighth with 443 patent families.
• TikTok owner ByteDance ranks ninth with 418 patent families.
• Software giant Microsoft ranks tenth with 377 patent families.

Source: WIPO, China-Based Inventors Filing Most GenAI Patents, WIPO Data Shows.

Our calculation: the top three Chinese applicants—Tencent, Ping An, and Baidu—cumulatively published 4,872 patent families, outnumbering the cumulative total of all top-ten Western corporate applicants (IBM, Alphabet, and Microsoft, totaling 1,421 patent families) by 3.4 to 1. This disparity occurs even as Western firms confront physical constraints in memory chip shortages and advanced hardware bottlenecks.

4. 37,000 inventions: The post-2023 commercial explosion

Any expectation that generative AI patenting would slow after the initial hype cycle was dismantled by WIPO’s mid-decade update. In its Technology SPARK analysis, WIPO reported that published GenAI patent families climbed from around 14,000 in 2023 to over 37,000 in 2025.

Over the two-year span of 2024 and 2025 alone, more than 56,000 new GenAI patent families were published globally. Our calculation: the volume of patent families published in 2024 and 2025 exceeded the entire cumulative patent output of the preceding decade (54,000 patent families from 2014 to 2023).

Source: WIPO, Patent Landscape Report and Technology SPARK Update.

Because patent applications are subject to an 18-month statutory publication lag between initial filing and public release, the surge in 2024 and 2025 reflects the tidal wave of enterprise R&D budgets mobilized immediately following the launch of commercial chatbot interfaces in late 2022.

Consequently, generative AI expanded its footprint within the broader intellectual property landscape. While GenAI represented 4.2% of all global AI patent publications in 2017 and 6.1% in 2023, its share jumped to 8.7% of all AI patenting in 2025.

5. 17,996 visual systems: Image, text, and molecular modalities

In analyzing the data modes utilized across generative AI inventions, WIPO revealed that visual systems dominated early patent filings. Between 2014 and 2023, image and video data accounted for 17,996 patent families.

Text generation followed closely with 13,494 inventions, while speech, sound, and music processing represented 13,480 inventions.

However, the composition of patent filings underwent a structural pivot between 2023 and 2025. According to WIPO’s SPARK update, text-related patent filings more than tripled in two years, closing the gap with image and video as multi-modal large language models expanded across enterprise workflows. Meanwhile, software code synthesis and 3D geometric modeling emerged among the fastest-growing categories, mirroring changing demands in global workplace task exposure.

The most specialized frontier is biochemical data. WIPO documented 1,494 patent families utilizing molecule, gene, and protein-based data between 2014 and 2023. While small in absolute numbers, molecular GenAI patents achieved a 78% average annual growth rate over five years. Generative systems are now actively patented for screening chemical compounds, predicting protein folding structures, and generating novel therapeutic molecules for targeted pharmaceuticals.

6. 56% annual growth: India and the emerging patent contenders

While China and the United States command the majority of headlines, the broader Asian continent is driving global patent momentum. The Republic of Korea ranks third globally with 4,155 patent families (an 8% world share), growing at an average annual pace of 38%. Japan ranks fourth with 3,409 patent families (a 6% share), led by conglomerate research at NTT and Sony Group.

The standout growth market is India. Ranking fifth globally with 1,350 patent families between 2014 and 2023, India achieved an average annual growth rate of 56%—the fastest among the top five inventor nations.

In Europe, the United Kingdom leads with 714 patent families (13% annual growth), closely followed by Germany with 708 patent families. Germany, however, demonstrated greater recent dynamism, posting an average annual growth rate of 30% as automotive suppliers like Bosch and industrial automation leaders staked intellectual property claims.

Together, the top seven inventor locations accounted for approximately 94% of all generative AI patent families published worldwide over the decade, reinforcing the sharp industrial output gap between India and China and the concentration of advanced digital capability among a handful of industrialized states.

7. 3-to-1 ratio: LLMs overthrow GANs in model architecture

Underlying these patent filings is a major architectural transition. During the first half of the decade, generative adversarial networks (GANs)—pioneered in 2014—represented the primary workhorse for synthetic media. Between 2014 and 2023, WIPO identified 9,700 GAN patent families, with 2,400 published in 2023 alone.

Variational autoencoders (VAEs) represented the second-largest historical category with approximately 1,800 patent families, while decoder-based large language models (LLMs) accounted for 1,300 patent families through 2023.

However, the technological hierarchy inverted rapidly. Between 2020 and 2023, LLM patent families expanded from 53 to 881, while diffusion models climbed from 18 to 441.

By 2025, WIPO’s SPARK report confirmed a decisive technological turning point: large language models officially overtook GANs as the dominant model type in global patenting, outnumbering GAN patents nearly 3 to 1 in 2025. Diffusion models rose to third place, cementing a permanent shift away from adversarial image training toward transformer-based auto-regressive generation and latent diffusion architectures, mirroring dynamics seen in the global clean technology patent competition.

8. 47,568 citations: The divergence between patent volume and research impact

A crucial nuance in international intellectual property analysis is the divide between patent quantity and scientific research influence. While China holds an unassailable lead in patent family counts, scientific publication impact tells a very different story.

Between 2010 and 2023, scientific research publications were evenly split between the two superpowers: Chinese institutions published 12,453 scientific papers on generative AI, while US institutions published 12,036 papers.

Yet when measuring research influence through scholarly citations, American institutions dominate by a wide margin:

• US-affiliated scientific publications received 169,693 global citations between 2010 and 2023.
• China-affiliated publications received 100,310 citations.
• United Kingdom institutions received 61,292 citations.
• Canadian institutions received 44,972 citations.

Our calculation: scientific papers from US institutions received 69.2% more citations than papers from Chinese institutions, despite producing a nearly identical volume of published research.

At the institutional level, Alphabet/Google is the preeminent global research powerhouse, generating 47,568 citations across its scientific publications—more than triple the citations of the top-ranked academic institution, the Chinese Academy of Sciences (15,674 citations).

The strategy of startup pioneers is equally revealing. OpenAI published only 48 scientific papers in the WIPO research corpus (ranking 325th by volume), yet those 48 publications accumulated an astounding 11,816 citations (ranking 13th globally). For years, OpenAI avoided patenting altogether, relying on trade secrets and open research dissemination before filing its first six US patent applications in early 2023.

How to read generative AI patent statistics

Navigating international patent databases requires understanding five vital methodological distinctions established in WIPO’s analysis:

1. Patent families reflect distinct inventions, not raw applications. Filing an identical patent in China, the United States, Europe, and Japan generates four national patent documents but only one simple patent family. Counting patent families prevents companies that file internationally from artificially inflating their output over domestic filers.
2. Statutory publication lags create an 18-month blind spot. Patent offices publish applications 18 months after their priority filing date. Consequently, patents published in 2025 reflect corporate research decisions and technical breakthroughs made in late 2023 and early 2024.
3. Patent volume does not equate to commercial dominance. Chinese patent filings are heavily encouraged by government subsidies, university graduation quotas, and domestic regulatory frameworks. A significant proportion of Chinese patents are filed solely at the China National Intellectual Property Administration (CNIPA) without international PCT filings.
4. Trade secrets compete with patenting. Software algorithms and AI model weights can be protected as proprietary trade secrets rather than published patents. Many leading AI labs intentionally withhold training methodologies and dataset curation techniques from public patent registries to avoid disclosing proprietary techniques to competitors.
5. Cross-industry convergence is accelerating. Early GenAI patents were dominated by internet search engines and social media networks. By 2025, industrial conglomerates (Bosch), electric grid utilities (State Grid Corporation of China), enterprise computing providers (Inspur Group), and semiconductor manufacturers (Nvidia) became major filing entities.

The global generative AI patent race at a glance

The following table synthesizes the core findings established across WIPO’s Patent Landscape Report and Technology SPARK Report, tracing the global evolution of generative AI intellectual property.

Sources: WIPO, Patent Landscape Report on Generative AI and Technology SPARK Report Update.

What to watch

• Cross-border patent disputes and licensing litigation, tracking whether Chinese tech firms attempt to enforce domestic patent claims against Western software providers operating internationally.
• The outcome of US and European patent office examinations regarding AI inventorship and the patentability of purely synthetic training routines.
• Patenting activity in biochemical and pharmaceutical design, observing whether GenAI-designed molecules patented between 2020 and 2024 successfully advance through clinical trials.
• Filing strategies of frontier labs including OpenAI, Anthropic, and xAI, evaluating whether they shift from trade secrets to defensive patent filing.
• Legislative and antitrust scrutiny regarding patent thickets, examining whether dominant enterprise holders leverage legacy patent portfolios to stifle open-source foundation model distribution.
• The acceleration of utility and infrastructure patents from new entrants like State Grid Corporation of China and Nvidia as generative AI embeds directly into electrical grids and physical edge silicon.

The bottom line

The global generative AI patent race reveals a profound divergence between scientific citation influence and intellectual property volume: while the United States dominates high-impact academic research and foundational model breakthroughs, China has constructed a massive 6-to-1 lead in physical patent filings, accumulating 38,210 inventions compared to America’s 6,276. With annual GenAI patent publications surging past 37,000 in 2025 and more than 56,000 new patent families published in 2024 and 2025 alone, the technology has transitioned from speculative experimentation into a permanent legal asset class. As large language models overtake GANs by nearly 3 to 1 and industrial heavyweights stake claims across telecommunications, energy, and medicine, intellectual property rights will dictate commercial leverage and licensing royalties for the next decade of digital automation. For ongoing coverage of how automated systems and computational breakthroughs are transforming global industries, visit our AI & Work section.

Related reading

• Stanford AI Index 2026 in 8 Numbers: The Capital Surge and the Shrinking US Lead: corporate investment flows, training compute benchmarks, and frontier model capabilities.
• Clean Energy Innovation in 8 Numbers: The Patent Race, Battery Dominance, and the $55 Billion Squeeze: clean energy patent filings, battery chemistry dominance, and global R&D investments.
• Generative AI Job Exposure in 8 Numbers: The ILO Global Workplace Index: occupational automation, clerical task displacement, and labor market structural shifts.