AI and Geopolitics

AI and geopolitics is the study of how states compete over, cooperate on and depend on artificial intelligence capabilities: national strategies, research and talent ecosystems, access to computing hardware, military and security uses, and the international institutions that attempt to govern them. It treats AI as an input to national power and economic position as well as a technology.

Frontier AI depends on inputs that are unevenly distributed across countries: capital, specialised chips, energy, data centres and skilled researchers (ECON-COMP-001). Governments therefore frame AI in terms of competitiveness and security, and several have published national strategies that set explicit goals. This article describes national positions, talent flows, compute access, security uses and dependence, and refers to other articles for detail that would otherwise be duplicated: export-control instruments are covered in GOV-EXPORT-001, and energy, data-centre and semiconductor supply chains in ECON-INFRA-001. The competitive framing ("race") is itself contested; it is used here only where a government source uses it.

China. On 20 July 2017 China's State Council issued the New Generation Artificial Intelligence Development Plan. According to a Stanford DigiChina translation, it set goals of keeping pace with leading AI levels by 2020, major breakthroughs in basic theory by 2025, and, by 2030, making China "the world's primary AI innovation center" (State Council of the PRC, 2017). On 26 August 2025 the State Council released opinions on an "AI+" action, which per the CSET English translation (Center for Security and Emerging Technology, 2025) were formulated on 21 August 2025 and set targets for integrating AI across the economy through 2027, 2030 and 2035 [VERIFICATION NEEDED: the translation was identified in search results but not opened; target figures and the term "penetration" are not quoted here].

United States. The White House released "Winning the AI Race: America's AI Action Plan" on 23 July 2025, listing more than 90 federal actions under three pillars: accelerating innovation, building American AI infrastructure, and leading in international diplomacy and security. It includes a commitment that the Commerce and State Departments will work with industry on full-stack AI export packages (hardware, models, software, applications and standards) for partner countries (The White House, 2025). The complete plan text could not be parsed in this session, so only the summary page is relied on.

European Union. The European Commission published the AI Continent Action Plan on 9 April 2025, describing it as a set of actions to make the EU a global AI leader while keeping trustworthy, human-centric AI (European Commission, 2025). The EuroHPC Joint Undertaking states that the EU "has established 19 AI Factories and 13 AI Factory Antennas" offering computing and support to firms and researchers; the same page lists countries that add up to 18 factories, so the count is inconsistent within the source (EuroHPC JU, 2026). Status of "AI gigafactories" was not retrievable [VERIFICATION NEEDED].

Multilateral. On 26 August 2025 the UN General Assembly adopted by consensus resolution A/RES/79/325, creating an Independent International Scientific Panel on AI of 40 members and a Global Dialogue on AI Governance. The Panel's scope is limited to the non-military domain. The Dialogue was launched on 25 September 2025, held its first annual session in Geneva on 6-7 July 2026, and a second session is scheduled for May 2027 in New York (United Nations, 2025; United Nations, 2026). See GOV-INTL-001.

Dimensions of national AI capability. A common decomposition, used here as an analytical device rather than an established index, is: (1) compute (chips, data centres, energy); (2) talent (researchers, engineers, and the ability to retain them); (3) data and deployment (domestic markets and public-sector adoption); (4) capital; (5) research output and models; (6) institutions (standards, regulation, procurement). No single published metric combines them, and composite rankings embed weighting choices that should be examined before use (ECON-MEAS-001).

Dependence. A state is technologically dependent on another where a critical input (for example advanced accelerators or cloud capacity) can be supplied, priced or withheld by that other state or by firms under its jurisdiction. Dependence is asymmetric and may be mutual: controls on chips affect the supplier's revenues, as shown by the supplier-side disclosures reviewed in GOV-EXPORT-001.

Strategic stability. In arms-control usage, stability refers to the absence of incentives to strike first or to escalate in a crisis. Whether AI affects stability, for example through faster decision cycles or uncertainty about an adversary's capabilities, is a matter of analysis and scenario work, not of measured outcomes; claims in this area are Hypothetical or Contested unless tied to a specific documented deployment.

Research ecosystems and national capability. The 2025 AI Index (first submitted to arXiv on 8 April 2025) reports that in 2024 US-based institutions produced 40 notable AI models, compared with 15 from China and 3 from Europe, and that performance differences between US and Chinese models on major benchmarks such as MMLU and HumanEval had narrowed from double digits in 2023 to near parity (HAI, 2025 web summary; Maslej et al., 2025). The 2026 edition's web summary states that US and Chinese models "traded the lead multiple times since early 2025" and that, as of March 2026, the top US model led the top Chinese model on one arena score by 2.7% (HAI, 2026). The same summary reports private AI investment in 2025 of USD 285.9 billion in the US and USD 12.4 billion in China, noting that the Chinese figure likely understates total Chinese spending. Differences in how "notable" is defined, and the exclusion of state-directed funding from private-investment measures, limit direct comparison.

Talent migration. The 2026 AI Index web summary states that "the number of AI researchers and developers moving to the U.S. has dropped 89% since 2017," including an 80% drop in the last year (HAI, 2026). The full methodology (the underlying data are reported by the Index to come from a professional-network dataset) was not read; the figure should be treated as Preliminary until the underlying chapter is checked [VERIFICATION NEEDED: data source, definition of "moving to" and base year]. The drivers (immigration policy, destination attractiveness, remote work) are not established by that source.

Compute access beyond the leading states. Public programmes such as the EU's AI Factories (EuroHPC JU, 2026) and the US plan's infrastructure pillar (The White House, 2025) are state responses to compute scarcity. The 2026 Index summary reports 5,427 data centres in the US, more than ten times any other country, and 85 public or public-private AI supercomputers in China (HAI, 2026). Chip export policy, which shapes who can obtain advanced accelerators, is described in GOV-EXPORT-001 and is not repeated.

National security. NATO published a summary of its revised AI strategy on 10 July 2024. It states four aims (lead by example in responsible AI use, accelerate adoption, protect AI technologies and manage their risks, and guard against adversarial use), endorses six Principles of Responsible Use (lawfulness, accountability, explainability, reliability, governability, bias mitigation), and flags AI-enabled disinformation as a risk (NATO, 2024). The Political Declaration on Responsible Military Use of Artificial Intelligence and Autonomy was unveiled in The Hague in February 2023 as non-legally binding guidance that does not create new obligations under international law, according to a West Point Lieber Institute summary; endorsement counts reported by secondary sources differ by date, and the official State Department page returned an access error (Lieber Institute, 2023) [VERIFICATION NEEDED: current list of endorsing states and text, from the State Department]. The UN Panel does not cover military uses (United Nations, 2026).

• Advantages of the framework: it highlights real constraints (chips, power, talent) and the policy instruments states use to shape them.
• Limitations: national-level comparisons rely on counts of "notable" models, on private-investment figures that exclude state funding, and on talent-flow data from commercial networks. Government plans state aims; this article does not assess their achievement.
• Unresolved: the extent to which US-China capability differences persist (the AI Index describes the benchmark gap as nearly closed on the metrics it uses, but benchmarks are imperfect measures; see CAP-BENCH-001); whether talent flows are a durable shift; whether "dependence" gives a state coercive leverage in practice; and whether military AI affects stability, for which no empirical record in the sources used exists.
• Framing risk: "race" language can encourage treating safety measures as costs, a concern raised in the governance literature (see GOV-INTL-001); no source for that claim was retrieved here, so it is attributed to the literature in general terms only.

As of 2026-10-15: (a) the US, China and the EU each maintain published AI strategies with different emphases (deployment and infrastructure in the US plan; industrial integration in China's plans; computing access and competitiveness in the EU plan); (b) the AI Index reports a US lead in counts of notable models and in private investment, and near parity between top US and Chinese models on its benchmark measure; (c) there is a multilateral scientific panel and dialogue under the UN with a civilian mandate; (d) military AI is addressed mainly through non-binding documents. Claims that one country "leads" are Contested unless the metric and date are stated.

Observed trends: continued public investment in compute, repeated use of trade and licensing policy, and development of multilateral scientific assessment. No quantitative forecast is registered in this article; dated developments belong in TIME-001 and indicators in FACT-001. Points to watch include the Panel's annual reports and the 2027 Dialogue session (United Nations, 2026), implementation of the EU plan, and updates to US and Chinese export-related policy (GOV-EXPORT-001). Scenario claims about strategic stability remain Hypothetical.

• Foundational Concepts:
  • ECON-COMP-001 The Economics of Compute and Scaling: why compute is a strategic input.
  • ECON-IND-001 The AI Industry Structure: the firms behind national capability.
• Related Methods & Technologies:
  • TECH-HW-001 AI Hardware: accelerators at the centre of dependence.
  • ECON-INFRA-001 Energy, Data Centers and Semiconductor Supply Chains: physical supply chains.
  • ECON-MEAS-001 Measuring the AI Economy: limits of cross-country indicators.
• Governance and Policy:
  • GOV-EXPORT-001 Export Controls and AI Geopolitics: chip and tool controls.
  • GOV-US-001 United States AI Policy
  • GOV-CN-001 China's AI Regulation
  • GOV-EU-001 The EU AI Act
  • GOV-INTL-001 International AI Governance and Compute Governance
  • APP-GOV-001 AI in Government and Public Services
  • CAP-BENCH-001 AI Evaluation: limits of benchmark comparisons.