IP News
Agentic AI Surge and Portfolio Divergence Signal New Enforcement Realities for Tech Filers
•
8 min read
•44 views
//
Worldwide artificial intelligence patent grants surpassed six figures for the first time in 2025, reaching 107,279 as filings for agentic AI rapidly accelerated. Data from IFI CLAIMS Patent Services highlights a growing strategic divide between traditional technology giants building massive patent portfolios and software-first developers relying on trade secrets. For corporate IP leaders, maintaining value in this shifting landscape requires rigorous claim drafting to navigate 35 U.S.C. § 101 subject-matter eligibility challenges.
According to the annual report IFI Insights: Inventing AI published on July 29, 2026, by IFI CLAIMS Patent Services, annual global AI patent grants reached 107,279 in 2025—an 83% increase over three years. Global AI patent applications totaled 209,518 in 2025. Generative AI accounted for 23% of these applications, while agentic AI represented 9%, up from 5% in the prior study period.
In the United States, overall AI patent grants declined by roughly 7% from 2023. However, subcategories showed strong growth: U.S. generative AI patents grew 11% year-over-year, and U.S. agentic AI patent grants rose 14% over the prior year and 70% since 2021. Globally, agentic AI filings grew 59% in the past year and 137% over two years, reaching 15% of total AI patent activity.
Samsung led global AI patent filings in 2025 with 2,162 applications, followed by Huawei (1,822) and Google (1,672). Samsung also ranked first in U.S. applications with 682 filings, slightly ahead of Google (671) and Microsoft (585). In generative AI, Google claimed the top spot globally and domestically. In agentic AI, Nvidia led worldwide and domestic filings with 225 global and 128 U.S. applications. Conversely, software-focused AI developers showed minimal patent activity: OpenAI filed 27 U.S. and 35 global applications in 2025, Anthropic filed seven U.S. and eight global applications, Alibaba filed four, while DeepSeek and xAI held no patents or applications.
Over the past decade, global AI patent grants have expanded at a compound annual growth rate of 35%, while applications grew by 29%. Despite this long-term trend, the U.S. market reflects an intentional shift in corporate IP strategy. IFI Senior Marketing Manager Lily Iacurci noted that companies are increasingly bifurcating their strategy—balancing robust patent portfolios, defensive filings, and trade secret protection based on the pace of AI evolution.
Furthermore, U.S. AI inventions face persistent statutory hurdles under 35 U.S.C. § 101. Software implementation claims that fail to emphasize specific technological improvements remain vulnerable to patent-ineligibility rejections during pre-grant prosecution and court invalidation post-grant.
The filing data reveals a stark tactical contrast. Asian conglomerates like Samsung continue to build massive patent moats across key jurisdiction markets, ensuring cross-licensing leverage and freedom to operate. In contrast, pure-play AI model creators frequently forgo patent filings to protect core algorithms as trade secrets, avoiding immediate public disclosure and subject-matter eligibility rejections.
For Korean and Asian filers appearing as applicants, relying heavily on broad, abstract AI claims creates elevated vulnerability under U.S. Section 101 standards. Conversely, when acting as defendants or opponents against competitor patents, strict eligibility standards benefit these companies by providing strong invalidation defenses. To balance these dynamics, prosecution counsel drafting AI patent applications should immediately implement a strict structural drafting practice: tie all software claims to concrete hardware architecture, specific algorithmic data pipelines, or tangible physical-system transformations (such as semiconductor execution optimizations or sensor-actuator interactions). This grounds the invention in technological application, hedging against abstract idea rejections.
As agentic AI transitions from software frameworks to embedded systems and autonomous decision agents, filing activity in this subcategory will continue to capture a larger percentage of total AI applications. Corporate IP managers should expect patent offices to scrutinize agentic workflows for functional claiming and patent eligibility.
While U.S. eligibility jurisprudence evolves, corporate IP owners should perform an immediate portfolio audit of pending AI applications. Organizations should identify software-only claims that lack detailed technical execution mechanics and proactively file continuation or divisional applications containing narrower, hardware-integrated or pipeline-specific claim sets before original applications issue.