Thirsty Algorithms : Regulating the water consumption of Artificial Intelligence (AI) in the European Union

Sammanfattning

This thesis examines whether the European Union's regulatory, policy, and transparency architecture for the water consumption of artificial intelligence (AI) infrastructure is fit for purpose, and tests the hypothesis that if good water-specific regulations on AI are adopted at EU level, then better water management of AI infrastructure will follow. The research uses desk-based document analysis of EU legislative texts (the AI Act 2024/1689, Energy Efficiency Directive 2023/1791, Commission Delegated Regulation 2024/1364, CSRD/ESRS E3, the Water Framework Directive, and the European Water Resilience Strategy adopted on 4 June 2025), European Environment Agency reports, the IEA's Energy and AI report (April 2025) and its April 2026 update Key Questions on Energy and AI, peer-reviewed literature (Li et al. 2023; de Vries-Gao 2025; Luccioni, Strubell and Crawford 2025), industry sustainability filings, and grassroots and journalistic sources. The empirical analysis is structured around Bacchi's “What's the Problem Represented to Be?” (WPR) framework, supplemented by the Jevons-Paradox lens developed by Luccioni and colleagues. The empirical evidence shows that AI water consumption is rising rapidly (de Vries-Gao (2025), in Patterns, estimates the global AI water footprint at 312.5 to 764.6 billion litres in 2025) while 30% of EU territory and 34% of its population experience water stress in any given year (EEA, Europe's environment and climate 2025). Current EU rules treat water as an ancillary metric of energy efficiency rather than as a substantive object of regulation. The AI Act contains no water provisions. The Energy Efficiency Directive's data-centre reporting captures water only above a 500 kW IT threshold and shields facility-level data behind commercial-confidentiality clauses. CSRD covers water comprehensively only for the largest undertakings and has been substantively narrowed by the 2025-2026 simplification omnibus packages. Comparative analysis of Germany's Energy Efficiency Act, the Netherlands' hyperscale prohibition, Ireland's grid-conditioned regime, and Spanish grassroots resistance (“Tu Nube Seca Mi Río”) shows that meaningful water governance is currently produced at sub-EU levels, not at EU level. I argue that the evidence is consistent with the hypothesis: better-targeted EU rules would plausibly improve water management, provided they treat water as a primary regulated quantity, through mandatory facility-level WUE disclosure, siting controls tied to Water Exploitation Index Plus thresholds, and binding water-positive trajectories, rather than as a secondary indicator of energy performance. Conceptually, the thesis situates this regulatory question within the planetary-boundaries framework (Richardson et al. 2023), which now identifies freshwater change as one of six of nine boundaries already transgressed, and within the governance literatures on multi-level governance (Hooghe and Marks 2003), polycentricity (Ostrom 2010), and the Brussels Effect (Bradford 2020). It concludes that the EU's current approach exemplifies what Neumayer (2003) calls weak sustainability, substituting efficiency gains and offset-style replenishment for absolute limits, and that a credible water-specific regime would need to move toward strong-sustainability instruments that hold AI infrastructure within the safe operating space of the basins that host it.