The Patent Ledger Says GenAI Went Industrial

GenAI ownership is migrating from Silicon Valley labs to industrial enterprises across Asia, and the patent ledger recorded the handover months before the market noticed: filings tripled in two years while investors kept staring at model benchmarks.

Why the consensus has the wrong frame

The consensus holds that patents are dead weight in AI. The argument sounds solid: the Alice doctrine renders abstract model claims unpatentable, trade secrets protect fast-moving assets better, and open-weight releases diffuse capability quicker than any 18-month examination queue. IP law firms spent 2025 telling clients that trade secrets are outpacing patents in AI portfolios. All of that accurately describes the model layer, and the model layer is a shrinking share of where GenAI value settles.

The application layer plays by industrial rules, and industry files patents. WIPO's July 2026 Technology SPARK update counts more than 37,800 GenAI patent families published in 2025, against roughly 14,000 in 2023. The combined 2024-2025 total, above 56,000 families, exceeds everything published across the entire 2014-2023 decade. GenAI's share of all AI patenting climbed from 4.2 percent in 2017 to 6.1 percent in 2023 to 8.7 percent in 2025. Filings now span finance, telecommunications, utilities and manufacturing. WIPO Director General Daren Tang put it plainly: GenAI patents now come from large enterprises in finance, telecom and infrastructure alongside the technology companies.

Look at who leads. SoftBank published close to 3,000 GenAI families, almost entirely in 2025: a deliberate, dated, single-year land grab. Ping An, an insurer, sits in the global top three. Six of the top ten applicants are based in China, and China-based inventors published over 43,000 families across 2024-2025, more than their own cumulative output for the prior decade. Japan grew filings at a 210 percent compound annual rate, Canada at 109 percent, the United States at 92 percent, Germany at 85 percent. The map of GenAI ownership and the map of GenAI market capitalization now show two different worlds.

The cost curve

Patents are a cost curve in disguise. Each family represents a priced bet, paid in drafting, filing and maintenance fees across jurisdictions, that a specific technique will be worth defending for twenty years. Read the volume as revealed corporate conviction: 2014: 733 GenAI families worldwide. 2017: 4.2 percent of AI patenting, the transformer's debut year. 2023: about 14,000 families, a 6.1 percent share. 2025: 37,800 families, an 8.7 percent share. That is a 51x volume increase in eleven years, with a 2.7x jump concentrated in the final 24 months.

According to AGORÀ Intelligence analysis of five sources, two primary WIPO datasets plus three independent corroborations, the composition shift matters more than the volume. The 2014-2023 wave was dominated by Chinese platform companies patenting model techniques. The 2024-2025 wave is dominated by application patents filed by enterprises that sell insurance policies, connectivity and machines. The cost curve says the industrialization phase started inside 2023 R&D budgets and surfaced in the 2025 publication data.

The cliff event

Patent families publish roughly 18 months after filing. The 37,800 figure for 2025 therefore encodes decisions taken in 2023 and early 2024, when enterprise GenAI budgets first hardened. The 2026 publication count, visible in WIPO's next update, will encode the 2024-2025 deployment boom, and every corporate curve in this dataset points upward.

The precedent is the smartphone. Handset patent filings surged from 2007 through 2009; the licensing and litigation wave landed in 2010-2012, once portfolios matured and market positions became worth defending. Solar tells the same story from the cost side: module prices fell 90 percent in a decade once industrial players took over from laboratories, and the patent record marked that handover too. Expect the GenAI equivalent in 2027: matured application-layer portfolios, the first cross-licensing frameworks, and the first serious enforcement actions between operating enterprises rather than between labs and publishers.

Three sectors that will look different by 2028

1. Insurance and finance. Ping An sits in the global top three of GenAI patent holders. Claims automation, underwriting copilots and reconciliation pipelines become licensable assets; mid-size insurers will license portfolios instead of rebuilding them, and IP position starts driving partnership choices.
2. Telecommunications. SoftBank's 3,000-family land grab targets network operations, agents and infrastructure GenAI. Carriers that filed early gain quasi-standard leverage over carriers that waited, echoing the handset dynamics of 2010.
3. Manufacturing. WIPO records GenAI filings spreading through industrial applications and utilities. Factory copilots, process optimization and design generation patents give early filers pricing power over late adopters, and Japan's 210 percent filing growth signals exactly where that pressure originates.

Kill signal: WIPO's 2027 update showing calendar-2026 GenAI publications below 38,000 families, flat against 2025, falsifies the industrialization thesis. A top-ten applicant table composed entirely of software and platform companies kills the ownership-migration claim.