Glossary · term

AI middle powers

AI middle powers are states or jurisdictions described as neither the dominant frontier-AI powers nor low-capacity followers, yet able to shape AI through some combination of technical capacity, industrial leverage, market size, regulation, diplomacy, or adoption. The phrase names an analytical middle tier, not an official legal or diplomatic class. Its membership depends on the author's criteria: some require the absence of a frontier-model developer, while others emphasize national capability, institutional strength, or influence.

Debate2022-12-06Wave 3 · 2025–26Maturity: 3/5

Origin and context

`Middle power` is an older international-relations category; adding `AI` has produced more than one taxonomy. Alex Etl's 6 December 2022 NATO study is the earliest exact English use directly verified here, placing Poland and the Netherlands below several European `AI great powers` in military capability. Anton Leicht's January 2025 essay used the phrase for most advanced economies outside the United States and China and proposed leveraging bottleneck industries. A 2026 peer-reviewed article then built an independent capability-and-status tier without citing Etl or Leicht.

Sources: s1, s2, s3

Why it matters

The label focuses attention on actors obscured by a United States–China binary. Depending on the analysis, their leverage may come from semiconductor supply chains, markets, research, standards, summit diplomacy, regulation, evaluation capacity, or adaptation of imported models. It can therefore help compare dependencies and strategic options. It does not imply that the countries form a bloc or should follow one roadmap: ECFR advocates pooled capability-building, while New America emphasizes adaptation and governance, both broader than Leicht's bottleneck prescription.

Sources: s2, s3, s4, s5, s6

Example

France shows why methodology must be stated. Etl treated France as an AI great power in a NATO military-capability analysis; Leicht and Blomquist later placed it among AI middle powers in United States–China-centered economic or status hierarchies. Neither label is simply a timeless fact about France. A careful sentence specifies the author, date, unit of analysis, indicators, and comparison set instead of presenting a permanent country list.

Sources: s1, s2, s3

How it differs

Sovereign AI

Sovereign AI is a strategy or capability objective concerned with meaningful control over AI dependencies. AI middle power is a relative state category. A state can pursue sovereignty while remaining classed as a middle power, and some middle-power strategies deliberately favor access or coalition-building over full-stack autonomy.

GPU-rich and GPU-poor

GPU-rich and GPU-poor compares actors' effective access to accelerators and infrastructure. AI middle power classifies states or jurisdictions using a wider mix of technical, economic, institutional, and geopolitical factors. Limited domestic compute may be evidence in one framework, but it is neither a synonym nor a sufficient test.

Compute Governance

Compute governance is a field of rules and technical measures concerning advanced computing resources. AI middle powers are possible subjects or authors of such policy, not a governance mechanism. The same jurisdiction may be influential in regulation while remaining dependent on foreign chips, clouds, or frontier models.

Maturity and evidence

Maturity is rated 3. The exact phrase is documented in 2022, received a distinct strategic treatment in 2025, and by 2026 appeared in peer-reviewed research, independent policy analysis, a multi-year convening, and a governance-mapping preprint. That is more than single-author circulation. Maturity 4 would overstate stability because sources still disagree about the defining metric, lower boundary, unit of analysis, and membership.

Sources: s1, s2, s3, s4, s5, s6

Limits and open questions

The category can smuggle value judgments into a seemingly technical ranking and may reproduce the exclusions it aims to analyze. National scores can hide private-company location, cross-border supply chains, unequal capacity within a country, or the EU's mixed role as bloc and set of member states. Frontier capability also changes quickly. Treat country lists as dated analytical outputs, not certifications, legal statuses, or forecasts, and keep descriptive classification separate from claims that one strategy will produce growth, autonomy, security, or influence.

Sources: s1, s2, s3, s5, s6

References

The Impact of AI on NATO Member States' Strategic ThinkingInternational Scientific Conference Strategies XXI · 2022-12-06 · class A
A Roadmap For AI Middle PowersThreading the Needle · 2025-01-23 · class B
Racing for recognition? Theorizing emerging status hierarchies and prestige competition in the AI eraInternational Affairs / Oxford University Press · 2026-05-11 · class A
Capability club: How the EU can lead the fight for AI middle powersEuropean Council on Foreign Relations · 2026-02-11 · class B
The Shangri-La Series: AI for Middle PowersNew America · 2026-06-29 · class B
Mapping General-Purpose AI Governance in Twenty AI Middle-Power JurisdictionsarXiv · 2026-08-18 · class A
Google Gemini Eats The World – Gemini Smashes GPT-4 By 5X, The GPU-PoorsSemiAnalysis · 2023-08-28 · class B

Last updated: 2026-09-07