Methodology
How the Patent Access Gap is measured: two official USPTO bulk data programs, counted in full, never sampled or modeled.
Summary
Every figure is computed directly from USPTO bulk files (PatEx and PatentsView) and re-verified against the raw data before publication.
Overview
How it works
The Patent Access Gap measures who the US patent system works for: who receives the grants, who abandons along the way, and how the AI wave is distributed. Every number is computed directly from official USPTO bulk data, the full base, never a sample, and no figure is modeled or extrapolated.
Grant side measures use USPTO PatentsView, research grade bulk tables covering every US patent granted since 1976 (9.45 million utility patents), with disambiguated assignees and full CPC technology classifications, current through December 31, 2025. Process side measures use the USPTO Patent Examination Research Dataset (PatEx), 2022 vintage, 14.1 million applications with entity size, status, and examination dates.
The two datasets have different clocks. PatentsView tells us who received patents through the end of 2025. PatEx tells us what happens to applications on the way, but its public snapshot ends in mid 2023, so process measures use mature filing cohorts (2005 to 2017 for grant rates, 2013 to 2018 for office action outcomes) where the outcome of nearly every application is known. Where recent cohorts are shown, the caveat says explicitly that the rates are floors.
Where a finding cuts against the narrative, it stays on the page: small entities that persist are granted slightly faster than large ones (median 762 versus 809 days to grant for the 2015 filing cohort). The barrier the data shows is surviving the process, not examiner speed.
Indicators
Indicators
Concentration (The 300 club)
Share of all utility patents granted in a calendar year taken by the top 300 disambiguated assignees. 2025: 42.6% of 325,822 grants; entry price 131 patents.
Inventor held share
Share of utility grants that are unassigned at grant or assigned only to natural persons, the standard USPTO proxy for independent inventors. 22.3% at the 1980 peak, 5.1% in 2025.
Grant rate by entity size
Share of applications eventually granted, by USPTO fee status (large versus small entity), for mature filing cohorts 2005 to 2017. The gap never drops below 14 points and reached 18.3 in the 2017 cohort.
Office action abandonment
Share of nonprovisional applications ending abandoned for failure to respond to an office action, pooled filing cohorts 2013 to 2018: micro 35.0%, small 26.2%, large 15.3%.
AI actor split
AI patents (CPC class G06N) by recipient type from disambiguated assignees. 2025: companies 12,809, universities 615, individuals 42, government 36; 305 company patents per individual patent.
Sources
Data sources
Both programs are official USPTO bulk data, downloaded in full and aggregated with SQL.
Caveats
Known limitations
- 01PatentsView assignee disambiguation can split corporate families into separate entities, so family level concentration is understated, never inflated.
- 02Entity size is USPTO fee status as last recorded, not company revenue; the micro entity category exists only from March 2013.
- 03Inventor held is a proxy: some patents unassigned at grant are assigned to companies later. The level is approximate; the fifty year trend is robust.
- 04Abandonment status text records that the applicant did not respond to an office action, not why: cost, discouragement, or a rational exit.
- 05AI is defined as CPC class G06N (core machine learning), a deliberately conservative scope that excludes AI adjacent classes.