Global AI Patent Surge: Trends, Dominance, and Strategy

For patent attorneys, few technology areas demand closer attention right now than artificial intelligence (“GenAI”). Two reports from the World Intellectual Property Organization (“WIPO”) bracket a two-year window: the Patent Landscape Report: Generative Artificial Intelligence published in 2024 (the “2024 Report”) and the Patent Trends Update in GenAI published in 2026 (the “2026 Report”). Together, they offer a data-driven view of how AI patenting has evolved and what that evolution means for innovators and the attorneys who advise them.

The 2026 Report highlights the accelerating pace of global AI patenting, a pace that commands the attention of attorneys advising technology-adjacent clients. WIPO’s 2026 Report updates the 2024 Report, documenting a field that has entered what it calls “a new phase of technological maturity and commercial urgency.”[1] GenAI inventions roughly doubled between 2024 and 2025, rising from approximately 18,862 published patent families to 37,808 total[2], more than 56,000 new GenAI patent families were published in 2024 and 2025 alone,[3] exceeding the entire cumulative output of the preceding decade.[4] GenAI now accounts for 8.7% of all AI patent activity, up from 6.1% in 2023. These are not speculative filings. While patent filings do not equate to enforceable IP rights, and many of these applications will face validity challenges before they mature into granted patents, the sheer volume reflects genuine trends in inventiveness. This is a technology area where companies are actively innovating, building IP portfolios, and signaling their intent to compete at the frontier.[5]

Geography Matters[6]

The geographic story is the one that matters most for strategic planning. China remains firmly at the forefront, with over 43,000 patent families published in 2024 and 2025 alone, more than its entire cumulative output over the preceding decade. But the growth is becoming more global. Japan has risen dramatically to third place among inventor locations, overtaking the Republic of Korea, driven largely by SoftBank’s extraordinary filing activity. The United States posted a 92% compound annual growth rate between 2023 and 2025, with patent families rising from about 1,185 in 2023 to over 4,380 in 2025. Among companies, the most striking shift is that SoftBank has emerged as the single largest GenAI patent holder with nearly 3,000 families, virtually all filed in 2023 and published in 2025. The traditional Chinese leaders, Tencent, Ping An, and Baidu, remain among the largest cumulative holders, while Alphabet has become the leading U.S.-based patent owner with 1,083 families. The sheer concentration of activity in East Asia signals that companies operating primarily in Western markets should be asking hard questions about their competitive positioning.

It is worth noting what the 2026 Report’s data reveals about the nature of these filings. International patent families, meaning patents filed in two or more jurisdictions, remain a small fraction of total GenAI activity, accounting for approximately 9% of all filings in 2025. The Report attributes this partly to the geographic composition of the landscape: Chinese applicants typically file most of their patents domestically, and the relatively young age of many GenAI portfolios means applicants have not yet extended protection internationally. This raises a forward-looking question that is underappreciated: as Chinese and Japanese companies mature their portfolios and begin nationalizing applications in commercially significant jurisdictions, companies that have not built competitive IP in overlapping technology spaces may face meaningful exposure through licensing demands, design-arounds, or outright freedom-to-operate challenges.

The Importance of Data Strategy[7]

According to the 2026 Report, “the most widely discussed GenAI trend of 2025 has been the rise of agentic artificial intelligence (agentic AI), or systems that can autonomously plan, execute multi-step tasks and adapt their behavior based on real-time feedback loops, with minimal human oversight.”[8] What sets agentic AI apart from earlier GenAI tools? Rather than simply responding to a single prompt, these systems break down complex goals into step-by-step actions, tap into external tools and data, and refine their own outputs along the way. The field is still young, but early patent filings are already emerging, with Alphabet (Google) and Nvidia among the first movers. As agentic AI matures, expect it to claim a growing share of the GenAI patent landscape.[9]

One practical consequence of this multimodal evolution is that the value of diverse training data becomes more critical, not less, as systems grow more capable. The 2026 Report documents how the emergence of highly efficient models from companies like DeepSeek has demonstrated that competitive reasoning performance can be achieved at a fraction of the costs previously assumed, while the broad release of open-weight models has democratized access to frontier-level GenAI. But having access to model architecture in the abstract is not the same thing as holding defensible rights to high-quality, domain-specific datasets suited for a particular commercial application, especially in life sciences and health care. Indeed, we may still be in something of a data gold rush: the computational infrastructure exists, the model architectures are increasingly commoditized, and the differentiator for any given AI application increasingly comes down to whether the developer holds access to proprietary, curated, and legally defensible training data. For patent counsel, this means that IP strategy cannot stop at the model layer. Clients need to think about how they protect and control the data pipelines that feed these systems. It is worth noting, however, that the legal tools for protecting training data remain imperfect. Trade secret protection, while often the most natural fit for proprietary datasets, presents novel enforcement challenges and requires planning to identify and protect the secrets well before they are misappropriated. Companies should consider approaching data protection as a layered strategy combining trade secret protocols, contractual restrictions, and technical access controls, recognizing that no single legal theory provides complete coverage.

The Life Sciences Story[10]

With respect to life sciences and health care, one of the most telling data points in the 2026 Report is how little space the sector occupies in an otherwise comprehensive analysis. The molecules, genes, and proteins data mode, which most directly captures life sciences innovation, showed uneven recent growth: 523 patent families in 2023, a dip to 397 in 2024, and a recovery to 766 in 2025. The Report notes that while this remains a relatively small category, it “continues to reflect the growing application of GenAI in drug discovery, protein design and the life sciences more broadly[11].” Looking forward, the Report observes that “the integration of GenAI into sectors such as healthcare, manufacturing, finance and energy is likely to expand the diversity of actors filing patents in this field.”[12] Patent activity in this space has not collapsed, but it has not kept pace with the explosive growth seen in text-based and image-based AI applications.

Life sciences AI patenting remains an area worth watching rather than a dominant theme in the current landscape. The connection to multimodal systems and data quality is direct and consequential for this sector: in healthcare and drug discovery, the stakes of flawed or biased training data are exceptionally high. As multimodal models increasingly integrate clinical records, imaging, genomic data, and literature within single architectures, the provenance and quality of each data source compounds in importance. For life science applications, where a biased algorithm could misdirect a drug candidate or misclassify a patient population, the quality and provenance of training data is critical.[13] While these risks are logical and well-grounded in the technical realities of multimodal AI, it should be noted that courts and regulators have not yet addressed data quality failures in AI-driven life sciences through enforcement actions or published liability findings. The legal theories under which such failures might give rise to actionable claims remain undeveloped, making this an area where the practical risk outpaces the legal framework.

Strategic Implications

For patent attorneys advising clients in this environment, strategic imperatives are straightforward but urgent. The 2026 Report states plainly that “the patent landscape is evolving faster than ever, and the strategic decisions being made now, about where to file, what to protect and how to navigate an increasingly crowded field, will shape competitive positions for years to come.”[14] Counsel should be benchmarking client portfolios against the competitive landscape described above, paying particular attention to whether claim language drafted around older GAN-specific architectures still captures the way innovation is actually happening now that LLMs and multimodal systems have become dominant.

Of particular consideration for U.S. innovators is the fact that patenting AI technologies is significantly more challenging in the United States than in many other jurisdictions due to the restrictions imposed by 35 U.S.C. § 101.[15] Under the Alice/Mayo framework, AI-related inventions face heightened scrutiny for patent eligibility, with claims frequently characterized as directed to abstract ideas implemented on generic computer hardware. The Federal Circuit has not drawn a clear line for AI method claims, and patent examiners apply the two-step eligibility test inconsistently. This creates an asymmetry: while Chinese, Japanese, and Korean innovators can secure patent protection for AI inventions under comparatively more permissive eligibility standards, U.S. applicants face a materially higher bar. Indeed, USPTO Director John Squires has acknowledged that “categorically excluding AI innovations from patent protection in the United States jeopardizes America’s leadership in this critical emerging technology.”[16] The result is a potential competitive disadvantage for U.S. innovators who limit their patent strategy to domestic filings. If the only patents in your portfolio are those that survived § 101 scrutiny in the United States, you may be leaving significant protection on the table in jurisdictions where those same inventions would be patentable without difficulty.

As a consequence, international filing strategy deserves fresh and urgent attention. While international patent filings remain only 9% of total GenAI filings, many portfolios have not yet been extended internationally. Filing in key jurisdictions before prior art density increases represents a genuine window. For U.S. companies in particular, global patent planning is not merely optional, it is a strategic necessity. Innovators who do not plan globally risk finding themselves unable to enforce against competitors in the very markets where AI patents are most readily obtainable, while simultaneously facing freedom-to-operate challenges from foreign portfolios that mature and enter Western jurisdictions. And for any client whose AI application depends on proprietary datasets, counsel should be advising on the full spectrum of protections available for that data, from trade secret protocols to contractual restrictions on downstream use, because in a landscape where model architectures are converging, proprietary data represents a significant competitive advantage.

[1] 2026 Report at Introduction

[2] .2026 Report at page 11.

[3] 2026 Report at page 20.

[4]  2026 Report at page 7.

[5] Id.

[6] 2026 Report at pages 13-16.

[7] 2026 Report at page 10.

[8] Id.

[9] Id.

[10] 2026 Report at 19.

[11] Id.

[12] 2026 Report at 20.

[13] See Abulibdeh et al., The Illusion of Safety: A Report to the FDA on AI Healthcare Product Approvals, PMC (2025) (finding that 95.5% of 691 FDA-cleared AI/ML devices lacked demographic transparency and only 1.6% had randomized trial evidence).

[14] 2026 Report at 8. See also, Congressional Research Service, Patent-Eligible Subject Matter Reform: An Overview, IF12563 (updated Jan. 16, 2026) (noting that “some stakeholders worry that AI inventions may be at risk under the current framework because they may be characterized as methods of organizing human activity, mental processes, or mathematical concepts”).

[15] See HGF, A Turning Point for AI Patent Eligibility? (2025) (comparing the EPO’s “technical character / technical effect” approach with the U.S. abstract-idea exclusion at the eligibility stage). 

[16] See https://news.bloomberglaw.com/ip-law/patent-chiefs-early-opinion-emphasizes-pro-ai-innovation-push