Seven components, each normalised to 0–100. The total is a weighted sum, and you set the weights. That is not decoration: the entire substantive difference between public AI-leadership indices comes down to weights, and hiding that would be dishonest.
These figures are not verified. The scores in this table are a placeholder for the structure of the ranking, not a measurement. Until they are reconciled with AI Index, Tortoise and OECD, neither the numbers nor the ordering mean anything. For that reason the link to the risk model is off by default.
Publications, citations, share at top conferences.
Applications, grants, share of the world total.
Researcher headcount, inflow and outflow.
Available compute, data centres, access to frontier chips.
Private and public funding, deal count.
Companies, revenue, penetration into the economy.
Regulatory maturity, evaluation institutions, treaty participation.
| # | Country | Score | Share of the leader |
|---|---|---|---|
| 1 | United States | 92.3 | |
| 2 | China | 76 | |
| 3 | United Kingdom | 47.3 | |
| 4 | South Korea | 44.6 | |
| 5 | Germany | 42.4 | |
| 6 | France | 39.7 | |
| 7 | Japan | 37.4 | |
| 8 | Canada | 36 | |
| 9 | India | 33.9 | |
| 10 | Israel | 32.7 | |
| 11 | Singapore | 30.9 | |
| 12 | Netherlands | 29.4 |
Race index
0.97
Herfindahl–Hirschman concentration of leadership across the top five. One means five equals, zero means a single dominant player.
The author's assumption, not an established fact: the tighter the race, the harder it is to agree. When on, the mitigation ceiling is multiplied by (1 − 40% × race index), currently −39%.