05 / OBSERVATORY · NUMBERS Back to home
The numbers of local and off-grid AI.
Ten numbers to understand why artificial intelligence is moving closer to data and closer to energy. Sources and dates next to each row.
Updated September 2026
Updated September 2026
415 → 945TWh · data center consumption worldwide 2024 → 2030
~20%projects at risk of delay from grid constraints
95 GWdata center connection requests in Italy
7–10 yrsgrid connection times in Europe
Consumption, grid, inference, sovereignty.
Each row carries its source and date.
| Figure | Value | Source |
|---|---|---|
| 01Electricity consumption of data centers worldwide | 415 TWh in 2024 → ~945 TWh in 2030 | IEA, Energy and AI |
| 02Data center projects at risk of delay from grid constraints | ~20% | IEA |
| 03Data center connection requests in Italy | 95 GW (531 requests); 1.9 GW in advanced stage; +48% in 2026 | Fortune Italia su dati Terna, ago 2026 |
| 04Grid connection times in Europe | up to 7–10 years | The Register, feb 2026 |
| 05Inference share of AI infrastructure spending | 55% in 2026, 59% in 2027 | Gartner via TechTimes |
| 06Companies running inference on private cloud (current or planned) | 56% versus 41% on public cloud | Broadcom Private Cloud Outlook 2026 (sondaggio di un fornitore) |
| 07Sovereign cloud spending | $80 bn in 2026, +83% in Europe | Gartner, feb 2026 |
| 08Energy for a single request to a model | ÷33 in 12 months (0.24 Wh per median Gemini prompt) | Google Cloud |
| 09People without internet | 2.2 billion | ITU, nov 2025 |
| 10People without electricity | ~730 million | IEA, WEO 2025 |