China files the most AI patents, but the United States shapes the technology through higher-impact inventions; Europe accounts for few AI patents and remains fragmented, with technological capability—not EU membership—driving cross-border AI knowledge flows.

Main Finding

Global AI patenting (2010–2023) is highly concentrated and increasingly capability-driven. China dominates in patent volumes, the United States leads in citation impact and technological influence, and Europe remains a fragmented, intermediate player: limited in volume, showing some signals of patent quality, but failing to function as an autonomous AI innovation pole. Cross-border AI knowledge flows are driven mainly by technological capacity and specialization; geographic or institutional ties (including EU membership) have limited independent effects once technological factors are controlled for.

Key Points

• Geographic concentration
  • China: 220,463 granted AI patents (2010–2023) — largest share.
  • United States: 80,371 patents, but dominates forward-citation impact.
  • Other sizable countries: South Korea (38,054), Japan (36,436), Germany (6,971).
  • EU aggregate: 16,689 patents — smaller than Japan despite more patentees; top five EU countries (Germany, France, Netherlands, Sweden, Ireland) produce ~83% of EU patents.
• Organizations and ownership
  • Most applicants are nationally based (~77% of firms), but multinationals (23% of firms) account for 56% of granted AI patents (230,058 patents), while national firms produced 44% (184,191).
  • Parent-company location matters: patenting activity often occurs in one country but is controlled by firms headquartered in U.S. or China.
• Sectoral composition
  • AI patenting concentrated in manufacturing (esp. computer/electronics), ICT, and professional/scientific/technical activities — these three account for ~67% of AI firms and ~60% of patents.
  • Education (universities/research centers): <3% of applicants but >18% of patents, highlighting the outsized role of academic research in AI inventions.
• Technological specialization and proximity
  • Measured technological proximity shows global convergence toward U.S. innovation trajectories; Europe’s countries are scattered rather than clustered into a coherent, autonomous trajectory.
  • Revealed comparative advantage (RCA) metrics indicate Europe (as a whole) is not relatively specialized in AI; smaller countries (e.g., Israel, Ireland) show stronger AI specialization.
• Knowledge diffusion (gravity model of bilateral patent citations)
  • The primary determinants of cross-border AI citation flows are technological capability and technological proximity/specialization.
  • Geographic, cultural, or institutional proximity (including EU membership) have secondary or negligible independent effects once technological variables are included.
  • Reported elasticity of AI patenting activity influencing citation flows is about 0.7 (reported in the paper), indicating technologically advanced and specialized systems are materially more likely to exchange AI knowledge internationally.

Data & Methods

• Core patent data: Stanford AI Index AI-patent dataset (derived from EPO PATSTAT), identified via a hybrid method combining keyword-based text analysis and classification-code approaches, validated by the AI Index team.
• Sample: Granted AI patents from 2010 through 2023, with bibliographic fields including authority, IPC/CPC codes, grant/publication dates, and forward citations.
• Firm linking: Patents merged with ORBIS Intellectual Property (BvD) to assign applicants, parent companies (control defined as ≥50.01% share), country of incorporation, NACE industry code, and firm attributes. Coverage caveats: ~17% of AI patents lack applicant data in ORBIS; 24 granted applications missing in ORBIS (<0.006%).
• Analytic measures:
  • Country- and EU-aggregate patent counts and citation-based impact measures.
  • Sectoral breakdown by 2-digit NACE and more detailed manufacturing sub-sectors.
  • Revealed Comparative Advantage (RCA) to assess national specialization in AI.
  • Technological proximity metrics based on patent content/IPC space to measure similarity of national portfolios to U.S./China trajectories.
  • Gravity-model estimation of bilateral patent citations to identify determinants of international knowledge flows (controls for economic/innovation size, technological proximity, geographic distance, cultural/institutional ties, EU membership).

Implications for AI Economics

• Capability-driven competition
  • Technological capacity and specialization are the main engines of cross-border AI knowledge flows; policy aiming to increase influence in global AI must prioritize building deep technological capabilities (R&D scale, talent, focused specialization) rather than relying primarily on institutional integration.
• Europe’s technological sovereignty agenda
  • Aggregating the EU does not automatically produce an autonomous AI pole. European policy should emphasize (i) coordinated investments to increase scale (R&D, cluster-building), (ii) strategic specialization where Europe can realistically lead, and (iii) mechanisms to translate high-quality research into patentable, scalable industry applications.
• Role of firms and multinationals
  • Multinational firms disproportionately shape patent volumes. European strategy can consider incentives for domestic scaling (to create national champions), attracting R&D headquarters of multinationals, and leveraging foreign-controlled local patenting to increase local knowledge spillovers.
• Universities and public research
  • The outsized patent contribution from education/research institutions highlights universities as leverage points for capability-building; policies that foster university–industry commercialization and scale-up could boost Europe’s patent footprint.
• International spillover modeling and policy evaluation
  • Gravity-style models that explicitly include technological proximity and specialization provide a more accurate picture of knowledge flows than models relying on geography or institutions alone. For economic research and policy evaluation, incorporating portfolio structure and capability measures is critical.
• Strategic trade-offs
  • Given global convergence toward U.S. trajectories, European policymakers face trade-offs between aligning with dominant external trajectories (to capture spillovers) versus investing to create complementary or niche capabilities that could sustain technological sovereignty without achieving parity in volume.

Limitations to keep in mind (relevant for interpreting results) - Patent counts are an imperfect measure of innovation (vary by sector, strategy, and propensity to patent). - ORBIS and patent-linking have coverage gaps (~17% applicant missing), and parent ownership assignment requires thresholds that may miss some control structures. - Citation-based quality measures are informative but subject to field- and country-specific citation practices.

Overall, the paper documents an increasingly bipolar AI patent landscape (China by volume; U.S. by influence) and shows that Europe’s pathway to technological sovereignty requires concentrated capability-building, larger-scale actors, and better integration of research, industry, and commercialization channels rather than relying on political integration alone.