Machinepower Index / Nations / France

Sovereign AI Score · Q4 2026 · Nation dossier

France

Strategic Autonomy. Build enough at home to keep the freedom to choose.

Score

6.62Rank 5 of 50 · Tier 2

Potential

7.10Rank 5 of 50

Lost to weakest link

−0.484th smallest gap of 50

Rank range

5–6Our rule, an average or a geometric mean

France is strongest in Will.Public trust and frontier model output hold it back.

France ranks 5th of 50 on the Machinepower Index with a Score of 6.62, and 5th on Potential at 7.10. It is in Tier 2.

Its weakest links are public trust at 4.5 (44th of 50, joint) and frontier model output at 5.2 (4th of 50, joint). Because the Score charges the weakest measures most, these cost it 0.48 points against its Potential, the 4th smallest gap of the 50.

Its strengths are public sector adoption at 9.7 and regulatory standing at 9.0. By layer it scores 6.5 on Watts, 7.3 on Weights and 7.5 on Will, against 50-nation medians of 5.0, 5.6 and 7.5.

On a plain average it would rank 5th; on a geometric mean, 6th. Our weak-link rule puts it 5th.

Assessment

France is positioning itself as the AI hub of Europe. With strong state backing for "Mistral" and nuclear-powered compute centres, it seeks a path to strategic autonomy that bypasses both US and Chinese dominance.

Leverage

No chokepoint, but the best siting position in Europe: nuclear power at scale is the continent's cheapest firm, clean electricity. France's leverage is deciding where European compute physically lands.

Watts, Weights and Will.Each the average of four measures.

Watts

6.5Median 5.0

Energy, grids, data centres and silicon: the physical capacity to run models at scale. Its lowest layer.

Weights

7.3Median 5.6

Frontier models, alignment capability, standards influence and the talent that produces them.

Will

7.5Median 7.5

The capacity to put AI to work in the state and the economy while keeping public consent.

Where the Score comes from.Each out of 10, ranked against all 50.

3 of the 12 are assessed against the written rubric rather than taken from a dataset. The rest are computed from public data. How each measure is set.

The twelve measures for France, each scored out of 10
MeasureOut of 10ScoreRank of 50
Watts · 6.5
Planning & PermittingAssessedHow quickly a nation can get large AI sites and their grid connections approved and built. 7.48th=
Firm Power CapacityHow much electricity a nation generates, and so how much compute it could power. 6.29th
Sovereign ComputeThe supercomputing capacity a nation holds on its own soil. 6.310th
Hardware & Silicon AccessWhether a nation can make, or reliably buy, the chips that AI runs on. 6.29th=
Weights · 7.3
Frontier Model OutputAssessedHow many leading AI models a nation produces. 5.24th=
Alignment CapabilityAssessedHow well a nation can test, evaluate and control advanced models. 8.24th
Regulatory StandingThe strength and influence of a nation’s AI rules and standards. 9.010th=
Talent DensityHow many of the world’s leading AI researchers and inventors work there. 6.77th
Will · 7.5
Policy CapacityWhether the state has the strategy, budget and institutions to act on AI. 7.834th
Public Sector AdoptionHow far government itself puts AI to work. 9.72nd=
Workforce TransitionHow ready a nation’s workers and labour policies are for AI. 8.023rd=
Public TrustWhether the public backs the build-out. 4.544th=
Score out of 10Weakest links50-nation medianAssessedJudged against the rubric, not taken from a dataset

The cheapest points are in the weak links.Close the gap, or trade around it.

Lift public trust and frontier model output to 6.5 and 6.2, and the Score rises from 6.62 to 7.11: 5th to 4th.

  • Public Trust+0.20
  • Frontier Model Output+0.13
  • Firm Power Capacity+0.08
  • Sovereign Compute+0.08
  • Hardware & Silicon Access+0.08
  • Talent Density+0.06

The Score gained from one more point on each measure, largest first. The rule weights the lowest measures most, so the cheapest gains sit in the weak links.

Who sits around it.Ranks 2 to 8.

RankNationScorePotentialWattsWeightsWill
2United States8.128.839.29.28.1
3South Korea7.507.777.77.28.4
4Japan6.807.037.56.47.2
5France6.627.106.57.37.5
6Germany6.557.176.87.17.6
7United Kingdom6.367.005.87.77.5
8India6.276.806.06.38.1

France trails Japan by 0.18 and leads Germany by 0.07.

Strategic Autonomy.One of eight strategic plays.

Strategic Autonomy

Build enough at home to keep the freedom to choose.

1 other nation in the Index plays it, and the EU, scored for reference, plays it too.

Want a written strategy for France: how it should play Strategic Autonomy and what to do next? Get in touch.

Built on the public record.Every measure, and where it comes from.

Epoch AIJuly 2026

European Frontier Model Weights

Mistral open-weight series

Mistral remains Europe's leading frontier-adjacent model developer in Epoch's continuously updated database, providing an independent open-weight core for European industry. It is the reason Frontier Model Output is scored 5.2 rather than the 3.0 that Epoch’s count of notable models alone would give.

View source
EuroHPC JUNovember 2025

Next European Exascale System

Alice Recoque (France) in pipeline

France will host Alice Recoque, Europe’s second exascale supercomputer after Germany’s JUPITER, under a contract signed in November 2025, and is among the countries competing for EU AI gigafactories, whose formal call was published in July 2026.

View source
Oxford InsightsJanuary 2026

Government AI Readiness Index 2025

#2 of 195 governments

France ranks second globally in the 2025 index, above the United Kingdom, supported by nuclear baseload energy for compute corridors and sustained state backing of the domestic AI ecosystem.

View source
IMF2023 data, published June 2024

AI Preparedness Index: human capital and labour market policies

0.162 of a possible 0.25 · 8.0/10 in the Index

Used for the workforce measurement. The IMF combines schooling, pay, labour mobility and active labour-market policies; the Index rescales it so the weakest of 174 economies scores 0 and the strongest 10.

View source
CSET Emerging Technology ObservatoryAugust 2026

AI safety research, 2019–2024

1,159 articles (2% of the world's) · 32.0 citations per article

A cross-check for alignment, which stays assessed. The Research Almanac counts English-language research only, so it understates Japanese, Korean and Chinese work.

View source
MeasureHow it is setSource
Planning & PermittingAssessed for France. Judged against a written rubric: how fast a nation can permit, connect and build large data centres and the grid behind them.Country evidence, checked against World Bank Business Ready 2025
Firm Power CapacityLog scale on total electricity generation.Ember and the Energy Institute Statistical Review
Sovereign ComputeLog scale on aggregate TOP500 performance.TOP500, June 2026
Hardware & Silicon AccessA three-part rubric: leading-edge fabrication tier, lithography access and export-licence status.Public fab, ASML and export-control records
Frontier Model OutputAssessed for France. Log scale on notable models released, with a stated uplift where credible domestic models fall below the notability bar.Stanford AI Index 2026 and Epoch AI
Alignment CapabilityAssessed for France. Judged against a written rubric: the depth of a nation’s safety, evaluation and alignment research.Country evidence, checked against the CSET Research Almanac
Regulatory StandingOxford Insights Government AI Readiness Index 2025, Governance pillar ÷ 10.Oxford Insights, 2025
Talent DensityLog scale on top AI authors and inventors, assessed where the dataset leaves a nation out.Zeki Data via the Stanford AI Index 2026
Policy CapacityOxford Insights Government AI Readiness Index 2025, Policy Capacity pillar ÷ 10.Oxford Insights, 2025
Public Sector AdoptionOxford Insights Government AI Readiness Index 2025, Public Sector Adoption pillar ÷ 10.Oxford Insights, 2025
Workforce TransitionThe IMF human capital and labour market policies sub-index, rescaled so the weakest of 174 economies scores 0 and the strongest scores 10.IMF AI Preparedness Index, 2023 data, published 2024
Public TrustThe share of adults who say AI products and services have more benefits than drawbacks, scaled so the most positive nation scores 10. Nations outside the survey are bridged from the KPMG and University of Melbourne trust study where it covers them, and otherwise set to the survey average of 55%.Ipsos AI Monitor 2026; KPMG and University of Melbourne, Trust in AI 2025

Score = 1 ÷ √(the average of 1 ÷ measure²) across all twelve. Potential = the average of Watts, Weights and Will. Q4 2026 edition, released 8 October 2026. The full method.

The Nowcast · From January 2027

France will not wait for the next edition.Neither does the Nowcast.

Every signal, the week it lands

Dated and sourced, tagged to the measure it moves, with a running estimate of France’s Score.

The written strategy

How France plays Strategic Autonomy, where its leverage lies, and what would change the picture.

Alerts and a weekly briefing

The day something moves a score you watch, and a written read every week.

See the Nowcast

£3,000 a year for the whole organisation. The first two weeks are free.