Machinepower Index / Dossiers / Italy
Italy
Diffusion Advantage — Import the frontier and out-deploy everyone. The returns of AI accrue to users, not only builders.
Score comes from the twelve measurements beneath the three layers. A nation needs all of them at once — world-class models are no use on a grid that cannot power them — so one weak measurement pulls the Score down instead of being averaged away. Potential is the plain average of the three layers. The gap between the two is what being lopsided costs: nothing if a nation is evenly built, more the further its strengths and weaknesses sit apart. The exact formula is on the methodology page.
The three layers
Energy, grids, data centres and silicon — the physical capacity to run models at scale.
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
Each layer is the plain average of its four measurements. The Score uses those twelve measurements directly rather than the three averages, because averaging hides things: a 10 next to a 3 comes out as a forgettable 6.5, and you lose both facts. Every measurement is scored 0–10 and marked to show where the number came from.
| Layer | Sub-dimension | Basis | Score |
|---|---|---|---|
| Watts | Planning & Permitting | Assessed | 4.8 |
| Firm Power Capacity | Dataset | 5.4 | |
| Sovereign Compute | Dataset | 8.1 | |
| Hardware & Silicon Access | Dataset | 5.8 | |
| Weights | Frontier Model Output | Dataset | 3.0 |
| Alignment Capability | Assessed | 6.4 | |
| Regulatory Standing | Dataset | 8.6 | |
| Talent Density | Dataset | 6.2 | |
| Will | State Procurement | Dataset | 6.9 |
| Public Sector Adoption | Dataset | 7.0 | |
| Workforce Transition | Assessed | 5.2 | |
| Public Trust | Dataset | 6.3 |
Dataset means the number comes straight from a published source. Assessed means we judged it ourselves, because no dataset measures that thing cleanly — 3 of the 12 here, and 116 of 312 across the Index. The CSV export flags them too.
Sources
Every source below was published within the last 12 months. These are the main ones behind Italy's profile. The full list, and the rule on how recent a source has to be, are on the methodology page.
The Bologna-based EuroHPC Leonardo system remains among the world's most powerful public research machines on the June 2026 list, anchoring Italy's national compute access.
View source →The Garante has taken direct enforcement action against AI services, including the ChatGPT proceedings, and remains among the most proactive regulators in Europe, within the EU AI Act framework now phasing in through 2026-2028.
View source →Italy's strategy focuses AI deployment on manufacturing and cultural-sector IP protection, leveraging shared EuroHPC infrastructure rather than sovereign frontier training.
View source →Bottlenecks
These hold Italy below its Potential. Its two lowest scores cost it the most: frontier model output at 3.0 (part of Weights) and planning & permitting at 4.8 (part of Watts).
Assessment
Italy focuses on the application of AI in its high-end manufacturing and luxury sectors. While it lacks sovereign compute, its "Made in Italy" AI strategy seeks to protect cultural and industrial IP.
The art of the possible
Italy's route is applied excellence: Leonardo-class public compute plus its manufacturing base could make it the world's benchmark for AI-transformed industry. It doesn't need frontier models — it needs deployment speed in its factories and ministries.
Strategic play — Diffusion Advantage
Import the frontier and out-deploy everyone. The returns of AI accrue to users, not only builders.
The play is editorial, not scored: it names the game Italy's measured position supports. Eight plays run across the Index — Total Sovereignty, Strategic Autonomy, Chokepoint Leverage, Diffusion Advantage, Compute Rentier, Specialist Power, Sovereign Capital and Siege Autarky.
Open Italy in the live Index → — every source behind every number, the comparison tool and the CSV export.