Machinepower Index / Dossiers / Japan
Rank 4–7 across the scoring rules tested. How this is tested
Japan
Chokepoint Leverage — For nations holding a step in the chain nobody can route around, which their government could switch off: trade access for compute, investment and protection, and stay irreplaceable.
How the Score, Potential and bottleneck cost are worked out: methodology.
The three layers
Energy, grids, data centres and silicon — the physical capacity to run models at scale.
Compute tracked by Epoch AI: Japan controls 18,000 H100-equivalents, wherever the sites are; has 600 on its soil with the owner not reported. Hosted compute is shown, not scored.
Models, alignment capability, standards influence and the talent that produces them.
The capacity to wire AI into the state and the economy while holding public consent.
Where the score comes from
| Layer | Sub-dimension | Basis | Score |
|---|---|---|---|
| Watts | Planning & Permitting | Assessed | 6.8 |
| Firm Power Capacity | Dataset | 7.0 | |
| Sovereign Compute | Dataset | 8.1 | |
| Hardware & Silicon Access | Dataset | 8.3 | |
| Weights | Frontier Model Output | Assessed | 5.0 |
| Alignment Capability | Assessed | 6.6 | |
| Regulatory Standing | Dataset | 8.0 | |
| Talent Density | Dataset Sources differ | 6.0 | |
| Will | Policy Capacity | Dataset | 7.0 |
| Public Sector Adoption | Dataset | 7.3 | |
| Workforce Transition | Dataset | 8.5 | |
| Public Trust | Dataset Sources differ | 5.4 |
Assessed marks a measurement we judged rather than took from a dataset: 3 of the 12 here.
Sources differ marks a cell where two sources disagree by more than a point. Talent Density: Stanford's count and the developer-based estimate differ by more than a point (5.2 against 6.9); the cell is their average. Public Trust: Ipsos and Melbourne/KPMG differ by more than a point (6 against 4.5); the cell is their average.
Sources
Share of adults agreeing that products and services using AI have more benefits than drawbacks; the Index scores public trust as 1 + share ÷ 10. Survey of 23,532 adults across 32 countries, March to April 2026.
View source →Japan has more than 30 major domestic LLM variants, several trained from scratch, including Preferred Networks' PLaMo, the National Institute of Informatics' LLM-jp-4, NEC's cotomi and SoftBank's Sarashina3. The Digital Agency selected seven domestic models for a government platform serving about 180,000 staff. None matches the US frontier on reasoning and code.
View source →Rapidus's 2nm pilot line in Chitose, Hokkaido has operated since April 2025, with successful gate-all-around 2nm transistor operation announced July 2025 and mass production targeted for H2 FY2027.
View source →The fully domestic ARM-based Fugaku (RIKEN) remains in the global top ten on the June 2026 list, serving public research while its successor programme advances.
View source →The June 2026 UK-Japan Frontier Technology Partnership pairs Japanese fabrication and investment with UK research depth across AI, semiconductors, quantum and fusion.
View source →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 173 economies scores 0 and the strongest 10.
View source →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 →Fossil, nuclear, hydro and other dispatchable plant; averaged with generation for firm power.
View source →Second source for talent, averaged with Stanford's count.
View source →Second source for public trust, averaged with Ipsos.
View source →Capital is shown beside the Score, not scored.
View source →Half of public trust: support for building more data centres, scored as 1 + share ÷ 10 and averaged with trust in AI.
View source →On the register — what the world cannot get anywhere else
Japan holds two of the register's chokepoints: positions nobody else can route around, which its government could actually shut off. See the whole chain.
Tokyo Electron holds about 92% of coater and developer track systems worldwide, and all of the EUV track. Japan put exactly this tool class under licence in July 2023. The ordinance controls equipment that coats, heats or develops resist using extreme ultraviolet radiation. The one alternative, Lam's dry resist, has a single announced customer after five years.
CSIS translation of METI Ordinance No. 25 →Japanese suppliers dominate the chemicals that lithography depends on. Tokyo restricted three of them against South Korea in 2019 (fluorinated polyimide, photoresist and hydrogen fluoride), which is what proved the chokepoint was real.
Japan METI →Bottlenecks
These hold Japan below its Potential. Its two lowest scores cost it the most: frontier model output at 5.0 (part of Weights) and public trust at 5.4 (part of Will).
Assessment
Japan is leveraging its legacy in robotics and high-end manufacturing to carve out a niche in "Edge Sovereignty." Its massive investment in domestic foundries (Rapidus) is a direct play for long-term material power.
The art of the possible
Rapidus reaching 2nm mass production would make Japan one of only three places on Earth fabricating leading-edge logic — transforming its position. Pair that with the software talent flowing from the UK partnership, and Japan's edge-AI and robotics dominance gets a frontier-grade foundation. Its talent-for-compute swap with India shows the method: trade what is abundant for what is binding.
Strategic play — Chokepoint Leverage
Stay the chain's sole chemistry set: keep the resist tracks and specialty chemicals ahead of any substitute, as Japan showed it could use them in 2019, and let Rapidus add leading-edge logic to a hand that already holds two keys. The weak layer is frontier models, and the chemistry is what buys access to them.
Leverage — the nearest position. Holds two chokepoints — the resist tracks and the chemistry — and is the only middle power to have proved one in use, against South Korea in 2019.
Open Japan in the live Index → — every source behind every number, and the comparison tool.