The generative AI boom has turned data centres from invisible utilities into one of the most land-, power- and water-intensive building types of the early twenty-first century. Planning theory, however, still tends to treat them through the dematerialising language of “the cloud.” This paper examines the institutional friction that the AI build-out has produced in Italy, where the national state and sub-national governments moved in opposite directions within a single quarter of 2026. Law Decree 21/2026 (Article 8) created a fast-tracked single national authorisation procedure (procedimento unico nazionale) with a binding statutory deadline of ten months, extendable to thirteen. Its purpose was to capture mobile hyperscale capital before it relocated to other European markets. Within weeks, the Lombardy region reasserted territorial control through Regional Law 11/2026, which routes large projects into supra-municipal planning agreements, requires wasteheat recovery audits, and imposes equalisation and compensation levies on built surface. We read these instruments against the actual reorganisation of Italy’s data-centre geography: the saturated Milan inference cluster, the brownfield conversion of the former Galileo Ferraris thermoelectric plant at Trino in Piedmont, and the subsea-cable and renewables story now unfolding as a “Data Center Valley” in Puglia. AI infrastructure, we argue, exposes a territorial blind spot in planning practice. We develop that argument through two distinctions: between the spatial logics of AI training and AI inference, and between data centres as engines of industrial requalification and as instruments of territorial extraction. The governance question is not whether to authorise these facilities. It is how to convert a highly automated, low-employment, resource-hungry industry, one that runs on accelerated corporate timelines, into durable local public value. That conversion requires a shift from reactive zoning toward proactive territorial equalisation.
Mehrdad Afshar Majd (Thu,) studied this question.