The off-grid AI observatory.
Numbers, models, energy and policy that explain why artificial intelligence is moving closer to data and closer to energy.
Seven reading paths through off-grid AI.
Each page collects figures, sources and dates, updated every quarter.
Numbers
Ten numbers on consumption, the grid, inference, sovereign cloud and access.
Models
The open models that run on site and how they are chosen.
Thematic models
Domain models for healthcare, public sector, research and industry.
Local languages
AI in the languages of the territory, beyond English: voice, translation, datasets.
Energy and AI
Why energy, not silicon, is the bottleneck.
Licenses
Open-source licenses, obligations and what it means to bring models on site.
Training
Courses, tools, sources to follow and events for anyone who installs, uses or studies open and local AI.
Notes from the observatory.
Short briefs on models, energy and policy. Published 27 September 2026.
Data centers in Italy: 95 GW waiting for the grid
Terna has 95 GW of data center connection requests, but only 1.9 GW are in an advanced stage. What it means for those who want on-site AI.
AI in African languages, even offline
Omnilingual ASR, WAXAL, Sunflower: in 2026 African languages gained open tools for voice and local AI. What you can use and under which license.
Inference overtakes training in AI spending
In 2026 inference is worth 55% of AI infrastructure spending and is moving to private clouds. Why it matters for those who want sovereign AI.