What a firm is
Each page is one legal entity, not a corporate group, so related entities appear as separate pages with their register relationships stated on each. A portfolio transferred between entities while applications were pending sits with the entity that received it, dated to the original filing years.
What is counted
Counts are of unique applications, keyed by application id; continuations and divisionals are separate applications and count separately. Each series is dated by its own event — filings by filing date, publications by publication date, grants by grant date. Applications have been published while pending only since 2001, and 91% of applications filed since then have been published; grants are complete through 2024 and publications through 2022, so more recent years are still filling in.
How you measure matters
A patent usually carries several CPC subclasses, so a technology table has to decide how a patent that spans three fields is shared among them. The footprint on each firm page shows the same patents counted three ways. Primary counting credits each patent once, to its first-listed class. Any counting — also called full counting — credits it once under every class it carries, so a patent with three classes appears three times and the column sums above the firm's patent total. Fractional counting splits each patent evenly, giving 1/n to each of its n classes, so every patent still sums to one.
Patent 10001950 carries subclasses G06F, H03M and H04L. Under primary counting it is one patent in G06F; under full counting it appears once in each of the three; under fractional counting it contributes one third to each.
Only fractional counting adds up: summed across every class it returns the firm's patent count exactly, while all-classes counting overshoots by the multi-class patents and primary counting hangs the whole of a shared patent on a single field. For that reason the technology tables order by the fractional column.
What counts as a classification: the invention's own inventional CPC allocations. The record carries 58.0 million allocations, and 34% of them are additional cross-references — secondary tags, not the invention's classification; the Y tagging schemes live almost entirely among them — and these are excluded from the counting conventions everywhere, on every firm page and in the national table alike. Without that screen, Y10 would rank as IBM's sixth-largest technology, and a Y code would reach the top-20 list of 80% of all firms. The screen unclassifies just 87 patents. The Y section itself — the tagging layer where nearly all those cross-references live (Y02 climate-change mitigation, Y04 smart grids, Y10 the USPC concordance) — is shown on the overview as its own table, counted as distinct patents carrying the tag. Separately, 928,874 granted documents carry no CPC classification at all — 852,418 design patents, 32,946 plants, 21,699 reissues and 19,043 utility patents — and appear in patent counts but not in any technology table.
Coverage
The panels hold 6,417,797 of the 9,361,444 granted patents in the source record, across 16,984 firms — those with at least 50 lifetime applications. Charts run 1980–2024.
Locations
Locations are read from the addresses on the document: a country for nearly every patent, and a city within that country for nearly all of those. Inventor addresses show where the people are; assignee addresses show where the patent was legally assigned, which for most firms is a head office.
A city reaches a map only if it survives the coordinate check: the point has to fall inside the country printed on the document, and the city text has to match the cluster the record disambiguated it into. 245 locations fail one of the two and are kept off every map. Their patents are not dropped — the country comes from the printed address and not from the point, so they stay in the national and firm totals; what goes is the county and the dot. The screen holds 5,470 patents off the county layer and withholds 324 firm-county pairs from firm footprints. 51 counties are left holding nothing else; they are drawn with no number and say so when opened — 1 on this map, and 50 in the island territories, which it does not draw at all.
Some activity is counted and never drawn, and Puerto Rico is most of it. 50 address clusters on the island, across 48 municipios, fail the coordinate check — every one the registry holds — so the county layer carries 0 municipios and the 958 patents naming an inventor address there sit in the United States country total, where the country is the printed one and the point never has to be trusted. A further 5 patents carry a raw Puerto Rico register code of their own, which no map draws. The world map likewise leaves out 46 country codes, carrying 6,431 inventor-patents, that the boundary file has no shape for — states that no longer exist (the Soviet Union, Czechoslovakia, Yugoslavia, East Germany) and register codes that were never countries. Both sit in the national totals and on no map.
Some addresses never reached a cluster at all. PatentsView disambiguates the addresses on a document into shared location records, and the geography here is keyed on those records, so an address it could not place had no key to be counted under — although most of them print a country plainly. 231,371 addresses of that kind are read back from the printed text and counted in their country, 98,233 patents between them; they carry no city and no coordinate, so they appear in country figures and on no map. Recovering them means reading the country codes the record uses, which include forms no longer current — the old codes for Japan, West and East Germany, England and Sweden among them — and those are resolved only where the cities filed under a code agree overwhelmingly on one country. Where they do not, the code is left unread rather than guessed: that is what happens to the codes for states which no longer exist, whose cities are now in several countries at once, and it leaves 6,882 addresses out of the country figures entirely.
Inventors
An inventor-year is a person named on any of the firm's filings that year, granted or published, deduplicated by PatentsView's inventor identity and dated by filing. When a firm holds both the published application and the granted patent of the same filing, only the grant is counted — the two records name the same people, and the grant's identification of them is the one kept.
Citations
Citations are prior-art references on the citing patent, largely supplied by the applicant rather than added by the examiner, and they are counted between firms in this panel only; firm pairs with fewer than 5 references between them are dropped.
A citation count is a measure of elapsed time before it is a measure of anything else. A patent can only be cited by patents granted after it, so every figure labelled citations received on these pages — the firm headline, the per-patent rate in the inventor tables, both flow tables — has not finished accruing for recent work, and the shortfall is severe. Averaged over all US patents by grant year, citations received run from 15.3 per patent in 1976 up to 34.8 in 1998, then fall away to 0.07 in 2025 — a 497-fold collapse from that peak. The fall is censoring and nothing else: a patent granted in 1998 has had 27 years in which to be cited, one granted in 2025 has had months. The earlier rise is a different matter, and mostly the growth of the later corpus that does the citing rather than anything about the patents being cited. No correction is applied for either. Compare firms of similar vintage, read a recent citation count as a floor rather than a quantity, and treat a firm whose patenting is mostly recent as understated against one whose patenting is mostly old.
Sources
PatentsView, built on the USPTO's published record, supplies the patents themselves and almost everything about them: grants and pre-grant publications, the inventors and assignees named on each, the CPC classifications, the prior-art citations between them, and the raw address text every map here is resolved from.
Bureau van Dijk Orbis supplies firm identity — which patents belong to one legal entity, what that entity is called, and how it sits in a corporate register.
The maps are drawn on US Census Bureau cartographic boundary files (2024) for counties and states, commuting zones dissolved from those counties on the USDA Economic Research Service 2000 crosswalk, and United Nations geodata for country outlines. Every county figure on this site is a point-in-polygon result against those boundaries, not a code copied from the patent record.