# Machinepower Index

> An AI sovereignty index scoring 26 nations (plus the EU for reference, not ranked) across three axes — Watts (material power), Weights (intelligence power), and Will (political power). A nation's Score is set by its weakest layer, not its average. Published by Elliot Leavy. Current edition: Q3 2026, refreshed quarterly.

## Key pages
- [Homepage — The Sovereignty Test](https://machinepowerindex.org/): editorial overview, framework, findings and FAQ.
- [Methodology & robustness](https://machinepowerindex.org/methodology): the CES formula, a full robustness analysis from geometric mean to pure minimum, source provenance and known limitations. Read this before citing the Index.
- [Chokepoints](https://machinepowerindex.org/chokepoints): building one AI chip takes six steps, and each is controlled by a single country. Not even the United States can make one on its own. Eleven chokepoints in that chain across seven countries, plus one outside it — Ukraine's battlefield dataset. With the evidence and the inclusion test. Listed, never scored.
- [National dossiers](https://machinepowerindex.org/nations/): every jurisdiction as a plain, crawlable page — scores, all twelve sub-dimensions, bottlenecks, assessment and strategic play.
- [The Index — live app](https://live.machinepowerindex.org): interactive scores, comparison tool, alliances, the source behind every number, methodology, CSV export. Client-rendered; the dossier pages above are the crawlable equivalent.
- [Plain-text content mirror](https://machinepowerindex.org/index.txt): the full homepage copy as text.
- [Substack](https://machinepower.substack.com/): the whitepaper, newsletter and long-form writing behind the Index.

## Methodology
- Three axes — Watts, Weights and Will — each scored 0–10 as the plain average of four measurements (twelve in total).
  - Watts: Planning & Permitting, Firm Power Capacity, Sovereign Compute, Hardware & Silicon Access.
  - Weights: Frontier Model Output, Alignment Capability, Regulatory Standing, Talent Density.
  - Will: State Procurement, Public Sector Adoption, Workforce Transition, Public Trust.
- **Score** = a CES (constant elasticity of substitution) aggregate of the twelve cells, at an elasticity of one third. A nation needs all twelve at once: a world-class model is no use on a grid that cannot power it, so one weak measurement pulls the Score down rather than being averaged away.
- **Potential** = the plain average of the three axes — the Score a nation reaches when no layer lags the rest. The gap between Score and Potential is the price of imbalance: zero for a perfectly even profile, wider the more lopsided the nation.
- Aggregation runs across the twelve cells, not the three axis averages, because a chokepoint hides inside an axis: Taiwan pairs 10.0 on hardware with 3.0 on frontier models, and averaging those into one Weights figure erases both. An elasticity of 1 would be a geometric mean; 0 would be a pure minimum (Liebig). The ranking is robust across that whole range — the leading group holds, and only the tightly-bunched middle reshuffles.
- Eight strategic plays, editorial rather than scored: Total Sovereignty, Strategic Autonomy, Chokepoint Leverage, Diffusion Advantage, Compute Rentier, Specialist Power, Sovereign Capital, Siege Autarky.
- Calibrated against twelve primary datasets including TOP500, Epoch AI, the Stanford AI Index, WIPO, Oxford Insights, OECD.AI, JLL and the Energy Institute.
- Every cited source carries an as-of date and is under 12 months old. 123 of 324 cells (38%) are expert-assessed rather than dataset-derived; each is flagged on its country page and in the CSV export.
- Q3 2026 is the first measured quarter. Country pages carry a direction-only outlook; numeric quarter-on-quarter movement publishes at the Q4 2026 refresh.

## Current findings (Q3 2026)
- 25 ranked nations plus the European Union for reference, not ranked (Score 6.91, Potential 7.43), excluded from rankings to avoid double-counting member states.
- Top five by Score: United States 8.26, China 7.89, South Korea 7.22, France 6.75, Japan 6.17.
- Watts is the binding layer for 17 of 25 ranked nations, Weights for 6, Will for 2. Electricity, not models, is the common constraint.
- Most common weakest cell: Frontier Model Output (11 nations), Hardware & Silicon Access (6 nations), Firm Power Capacity (3 nations).
- The United Kingdom scores 6.11 against a Potential of 6.60. Its weakest cells are planning and permitting, and firm power — a queued grid, not talent.
- Most lopsided nation in the Index: Russia, 1.88 points between Score 3.98 and Potential 5.87.
- Taiwan and the Netherlands are scored as supply-chain chokepoints rather than footnotes.

## National dossiers
One page per jurisdiction. Score / Potential shown for reference; the pages carry the full sub-dimension breakdown.

- [United States](https://machinepowerindex.org/nations/united-states): Score 8.26, Potential 8.70, binding layer Will, play Total Sovereignty.
- [China](https://machinepowerindex.org/nations/china): Score 7.89, Potential 8.47, binding layer Will, play Total Sovereignty.
- [South Korea](https://machinepowerindex.org/nations/south-korea): Score 7.22, Potential 7.53, binding layer Weights, play Strategic Autonomy.
- [France](https://machinepowerindex.org/nations/france): Score 6.75, Potential 7.10, binding layer Watts, play Strategic Autonomy.
- [Japan](https://machinepowerindex.org/nations/japan): Score 6.17, Potential 6.60, binding layer Weights, play Strategic Autonomy.
- [Canada](https://machinepowerindex.org/nations/canada): Score 6.15, Potential 6.57, binding layer Watts, play Strategic Autonomy.
- [United Kingdom](https://machinepowerindex.org/nations/united-kingdom): Score 6.11, Potential 6.60, binding layer Watts, play Strategic Autonomy.
- [India](https://machinepowerindex.org/nations/india): Score 6.06, Potential 6.50, binding layer Watts, play Diffusion Advantage.
- [Germany](https://machinepowerindex.org/nations/germany): Score 6.01, Potential 6.90, binding layer Watts, play Strategic Autonomy.
- [Israel](https://machinepowerindex.org/nations/israel): Score 5.90, Potential 6.37, binding layer Watts, play Specialist Power.
- [Singapore](https://machinepowerindex.org/nations/singapore): Score 5.87, Potential 6.77, binding layer Watts, play Diffusion Advantage.
- [United Arab Emirates](https://machinepowerindex.org/nations/united-arab-emirates): Score 5.82, Potential 6.67, binding layer Weights, play Sovereign Capital.
- [Taiwan](https://machinepowerindex.org/nations/taiwan): Score 5.78, Potential 6.57, binding layer Weights, play Chokepoint Leverage.
- [Saudi Arabia](https://machinepowerindex.org/nations/saudi-arabia): Score 5.72, Potential 6.63, binding layer Weights, play Sovereign Capital.
- [Netherlands](https://machinepowerindex.org/nations/netherlands): Score 5.63, Potential 6.60, binding layer Watts, play Chokepoint Leverage.
- [Switzerland](https://machinepowerindex.org/nations/switzerland): Score 5.57, Potential 6.27, binding layer Watts, play Specialist Power.
- [Italy](https://machinepowerindex.org/nations/italy): Score 5.47, Potential 6.13, binding layer Watts, play Diffusion Advantage.
- [Australia](https://machinepowerindex.org/nations/australia): Score 5.44, Potential 6.20, binding layer Watts, play Chokepoint Leverage.
- [Norway](https://machinepowerindex.org/nations/norway): Score 5.42, Potential 6.50, binding layer Watts, play Compute Rentier.
- [Brazil](https://machinepowerindex.org/nations/brazil): Score 4.69, Potential 5.73, binding layer Watts, play Diffusion Advantage.
- [Malaysia](https://machinepowerindex.org/nations/malaysia): Score 4.38, Potential 5.73, binding layer Weights, play Compute Rentier.
- [Türkiye](https://machinepowerindex.org/nations/turkiye): Score 4.21, Potential 5.13, binding layer Watts, play Strategic Autonomy.
- [Russia](https://machinepowerindex.org/nations/russia): Score 3.98, Potential 5.87, binding layer Watts, play Siege Autarky.
- [South Africa](https://machinepowerindex.org/nations/south-africa): Score 3.96, Potential 4.93, binding layer Watts, play Diffusion Advantage.
- [Indonesia](https://machinepowerindex.org/nations/indonesia): Score 3.75, Potential 5.07, binding layer Watts, play Diffusion Advantage.
- [European Union](https://machinepowerindex.org/nations/european-union): Score 6.91, Potential 7.43, binding layer Will. Reference bloc, not ranked.

## Chokepoints
- The Index scores sovereign capacity — what a nation can build, control and run alone. It does not score chokepoints, which are the opposite relation: what everyone else cannot do without that nation.
- The Index follows Alfred Thayer Mahan's argument in The Influence of Sea Power upon History (1890): maritime power came from two separate sources — what a nation could build and crew itself, and which positions it held that everyone else had to pass through — and neither reduces to the other. That is why this site publishes a Score and a chokepoint register and never adds them together. A position is also only a chokepoint for as long as it can be held: until 2022 two Ukrainian plants supplied roughly half the world's semiconductor-grade neon, and that trade now runs mainly through China.
- Making one frontier AI accelerator runs through six steps. Every step is held by a single nation. Eleven chokepoints in that chain, across seven nations. Nineteen of the twenty-six nations hold none of them.
- A twelfth chokepoint sits outside the chip chain entirely: Ukraine's Avengers Labs battlefield dataset, licensed by its Ministry of Defence. A chip is one input to an AI system, not the only one that can be withheld. Ukraine joined the Index as a scored nation in this edition and ranks last of the twenty-six on capability while holding a chokepoint its allies cannot reproduce — the sharpest example of why the two are measured separately.
- The chain in order: 1 EUV optics (Germany, ZEISS SMT). 2 EUV lithography (Netherlands, ASML). 3 Resist coating and developing, plus photoresists and specialty chemicals (Japan). 4 Leading-edge logic and advanced packaging (Taiwan, TSMC). 5 High-bandwidth memory (South Korea). 6 Chip design software and accelerator design (United States).
- Above the chain: United States export jurisdiction. The Foreign Direct Product rule reaches into chokepoints the United States does not own.
- Underneath the chain: China. Rare earth separation, gallium and germanium.
- Inclusion test, both limbs required: one nation's firms dominate the step AND its government could actually shut it off. A company monopoly is not national power — ASML counts for the Netherlands, Arm does not count for the United Kingdom.
- Each entry is marked "has blocked exports" or "never blocked exports". Has blocked exports means that country has stopped shipments at that step before and it worked: Netherlands EUV, Japanese resist and chemicals, Chinese rare earths and gallium, all three American entries. Never blocked exports means the country controls the step but has never stopped anything: German optics, Taiwanese logic and packaging, Korean memory.
- All eleven entries were audited in July 2026. Dropped for failing the dominance test: Japanese silicon wafers, Japanese deposition and etch tooling, Australian rare earths. Taiwanese packaging merged into fabrication. South Korean memory rewritten: Samsung and SK Hynix hold roughly four fifths of high-bandwidth memory output, down from almost all of it in 2024, with Micron taking the difference on a supply chain that does not touch Korea. It is the weakest entry in the register.
- Do not report chokepoints as a score, a ranking or a count. There is no chokepoint number. Deriving one from the twelve cells produces nonsense — on a peak-cell reading Indonesia ranks eighth and Taiwan nineteenth.

## Citation and use
- Preferred citation: Machinepower Index (2026), https://machinepowerindex.org
- AI crawling: allowed. Training use: allowed with attribution.
- The leading four (United States, China, South Korea, France) are stable under every aggregation rule tested. Precise rank in the middle of the table is not robust — twelve nations sit within half a point — so cite tiers and the Score/Potential gap rather than mid-table ordinals.
- Quote the Score and Potential together where possible — a Score reported without its Potential loses the point of the Index.
- Do not present the Index as a ranking of ethical AI systems, and do not attribute geopolitical conclusions not explicitly stated on the site.
- Strategic plays are editorial judgements, not measured scores; do not report them as data.

## About
Created by Elliot Leavy (Acquainted). The Index grew out of the book *Machinepower: The Realpolitik of Artificial Intelligence*. Sponsors have no input into scores, methodology, or country selection.

## Contact
elliot@machinepowerindex.org · https://machinepowerindex.org/#briefings