01 Five years.
324,040 patents.
Twenty companies.

This page reads U.S. utility-patent grants for 2020-2024. The source is the USPTO Open Data Portal - PatentsView Granted Patent Disambiguated Data (product PVGPATDIS), whose bulk tables cover 1976 to the present. All counts are granted, non-withdrawn utility patents; a patent's year is its grant year.

Data status

Every figure on this page is real, computed directly from the PatentsView bulk tables by scripts/build-data.ts: it streams g_patent.tsv, g_assignee_disambiguated.tsv, g_cpc_current.tsv, and the disambiguated inventor + location tables through DuckDB and writes src/data/derived.json. Grant counts, per-year trajectories, shares, CPC-section leans and each firm's portfolio mix are all measured from the raw grants. Only a small set of labels stays curated - assignee family display names (a disambiguated cluster is shown under its recognizable name), geography anchor employers, the sector-table titles and examples, and the compare-chip short names.

Trajectories (slope graph)

The slope graph shares one vertical scale across all 20 lines. Four series are highlighted in red: IBM (-73% over five years, the dramatic collapse from a three-decade #1), TSMC (+41%, the steepest climb among the fabs), and two Asian challengers, Huawei (+10%) and Hyundai Motor (-12%). The remaining sixteen are rendered in light grey for shape, not identification. To avoid a pile-up, endpoint names are shown only for the four movers plus the largest steady filers (Samsung, Qualcomm, LG); on narrow screens the names are dropped entirely. Every company is named in the small-multiples grid.

Small multiples

Each plot rescales to its own y-axis so trajectory shape reads regardless of magnitude. The five-year delta in the upper right is unscaled. Companies with a five-year decline beyond 5% are tinted red; rises beyond 5% are inked in black.

Sector composition (coxcomb)

Wedges are equal-angle, with wedge area proportional to the section's share of all 2024 grants (so radius scales with the square root of share - the honest Nightingale construction, where the eye reads area). The ranked list beside it carries the exact percentages and counts and is the primary read. Section codes are top-level CPC: H Electricity, G Physics, A Human Necessities, B Operations & Transport, C Chemistry, F Mechanical, E Constructions, D Textiles. A patent counts toward every section it's classified under; here we normalize to the dominant classification per patent.

Compare

On the home page the fixed Samsung × IBM pair leads with the parallel-bars table - reading toward the centerline you see Δ in percentage points, with positive numbers favoring assignee A, negative favoring B - above a trajectory overlay that isolates the same two series on a shared scale. The compare page generalizes this to any two of the top 20 assignees with a fully static CSS-only picker (no JavaScript): because an arbitrary pair is chosen in the browser, no pairwise Δ is computed there - instead the trajectory, 2024-share, and CPC-mix marks all use one shared scale in both panels, and the mix bars mirror outward from the centerline, so a mark's size is the comparison across the gutter.

Out of scope

Citation counts (which patents cite which) - the most-cited inventor list is more interesting than the most-prolific list, but citations require g_us_patent_citation.tsv, which is the largest bulk file and is not yet ingested here. Pre-grant publication data, design / plant patents, and reissues are also out of scope. The compare picker is live over the top 20 assignees; a future pass can add a citation-impact dimension (and inventor / co-inventor networks) as a fourth axis alongside trajectory, share, and CPC mix.