Mastercard’s patent mix has started to tilt toward agentic payments. Visa’s has not moved
Patent publications fell 14.1% in the 10 months to July 2026 against the same period a year earlier, counted by publication date to keep the comparison like for like. The set is Visa (NYSE: V), Mastercard (NYSE: MA), Capital One (NYSE: COF), JPMorgan Chase (NYSE: JPM), PayPal (Nasdaq: PYPL), and Block (NYSE: XYZ), chosen to span the main business models in US payments: two card networks, a card issuer, a universal bank, and two fintechs. It is not a ranking of the sector’s largest patent filers, and American Express, Fiserv, FIS, and Global Payments sit outside it.
Most of the fall belongs to two companies. Capital One and JPMorgan Chase account for 73% of it, and Capital One alone produced 41% of the group’s current-year output, more than the two networks combined. The contraction sits in two large in-house bank portfolios rather than in the card networks. Publication counts also have no market baseline here, so part of the fall may reflect patent-office throughput rather than any decision a company made. What the data supports is a comparison between companies.
Figure 1: Only Mastercard and Block increased patent publications; the other four fell by 15% or more
Source: GlobalData Patent Analytics
AI language is rising while AI classifications shrink
Publications whose titles carry explicit generative AI or AI-agent language, such as large language models, foundation models, prompts, multi-agent frameworks, and agentic systems, rose from 31 to 104. Their share of total output moved from 0.6% to 2.4%.
The classification data complicates that picture. Publications carrying any artificial intelligence or machine learning classification fell 24.4%, and the generative-model subgroup was flat. The AI block inside these portfolios is shrinking. What changed is which problems these companies are willing to describe as generative or agentic in the public record, making the title-level shift a read on intent rather than on scale.
The shift is also concentrated. Generative and agent disclosures now account for 7.4% of PayPal’s publications, 5.0% of Mastercard’s, and 3.4% of JPMorgan Chase’s, against 1.6%, 0.7%, and 1.1% respectively a year earlier. Capital One sits at 1.0%. Visa sits at 0.7%, against 0.5% a year earlier. Block, the only other company to grow its publication count, recorded none.
Figure 2: Mastercard’s generative AI and agent disclosures climbed to 5.0% of its publications while Visa’s barely moved
Mastercard added clusters without losing volume
Mastercard is the only company in the set that did both. Its publications rose 2.2%, and 6.3% once patents inherited through acquisitions such as NuData Security, Vocalink, and CipherTrace are excluded.
Three clusters appeared from almost nothing. Open banking titles went from none a year earlier to 25, or 4.5% of Mastercard’s output, and several of them combine open banking infrastructure with large language models. Quantum-related titles rose from one to nine, including quantum-resistant offline authentication for payment devices. And one application, “Agentic microtransactions through verifiable payment delegation,” is the only disclosure across all 9,403 publications in both windows whose title uses the word agentic.
Mastercard also pushed protection outward. Publications filed through the World Intellectual Property Organization and the European Patent Office rose from 27.7% to 33.3% of its total, while its US share fell from 59.4% to 53.5%.
These clusters sit in open banking, data services, and agent-initiated payment authorisation rather than in the switching stack. That is where any commercial follow-through would show up first.
Visa’s mix did not move
Visa went the other way on volume. Its publications fell 18.4%, and 16.2% once the Featurespace acquisition is excluded, with US publications down 27.8%.
The composition barely changed. Two in five of Visa’s AI-classified publications sit in transaction authorisation, security, or fraud work: 40.4% this year against 40.2% last year. Its generative-model classifications moved from six publications to nine, and one 2025-priority disclosure covers foundation model fine-tuning. The portfolio Visa disclosed this year looks like the portfolio it disclosed last year, only smaller.
Mastercard has not abandoned risk work either. It runs a higher share of its AI in authorisation and security than Visa does, at 53.4%. The difference is that Mastercard added new clusters on top of that base, and Visa did not.
One finding runs against the prevailing narrative. Blockchain and cryptocurrency-tagged publications fell 27.3% across the group. The two networks account for 73% of that decline between them, with Mastercard’s own blockchain-tagged output down 34.9% and Visa’s down 36.5%. Across more than 9,000 publications in both windows, no title contains the word “stablecoin,” including in the year after the GENIUS Act established a federal stablecoin framework in July 2025.
What the data cannot settle
These are publications, not filings, and 72% of the frontier publications carry 2024 priority dates. Filings made after the agentic-commerce announcements of spring 2025 will not surface in volume until late 2026 and through 2027. The title screen is a floor rather than a census.
Three things would test the reading. The late-2026 and 2027 publication windows are the first to carry 2025-priority filings, and the first useful check on whether Visa patented its agentic-commerce work or left it unpatented. Mastercard’s applications rose while its grants fell, so its pending pipeline is building and the next 12 to 24 months will show whether it converts. And any appearance of stablecoin or tokenised-settlement language would date the industry’s real filing response to the GENIUS Act.
Patent activity is one thing. Whether it is investable is another
GlobalData’s Extracting Innovation Alpha Using Patents tests that question on the S&P 500 from 2019 to 2025, using more than 40 proprietary patent indicators. Explore the full analysis, including annualised alpha of between 5% and 7% over the S&P 500 and between 6% and 9% over the ARK Innovation ETF, the NYLI US Large Cap R&D Leaders Index, and the Solactive Global Innovation Index; the generative AI leadership matrix, in which patent-identified leaders outperformed laggards by 1.7 times; and the Brinson attribution separating stock selection from sector allocation.
Download for free below.
