>>Artificial Intelligence and Environment/Sustainability
The introduction of Artificial Intelligence (AI) into environmental and sustainability fields is often presented as an inevitable technological solution to the climate crisis. Indeed, AI already plays a key role in several environmental applications; it supports extreme-weather prediction, renewable-energy forecasting, wildlife monitoring, and biodiversity assessment (Vinuesa et al., 2020).
These examples show that AI can provide genuine operational value in specific domains. However, many of the more ambitious sustainability applications, such as improving energy grids, making infrastructure climate-resilient, or reducing industrial emissions with AI, remain largely at the research or pilot stage, with limited evidence of widespread deployment (Johnson et al., 2025; Rohde et al., 2024). Even where AI tools are deployed, they typically address well-defined tasks rather than delivering the broad, systemic transformations often invoked in public narratives. Distinguishing between established applications and those that remain emergent or speculative is therefore essential for understanding AI’s real contribution to environmental and sustainability goals.
Critical Environmental Problems and Risks
Epistemic Blindness in AI Narratives: Public and policy discourse often frames AI as an inevitable climate solution while obscuring the energy-intensive computing infrastructure, extractive supply chains, and e-waste streams that sustain it. This mythologising framing creates a collective blind spot around AI’s ecological footprint (Brevini, 2020; Crawford, 2021).
Escalating Computational and Energy Demand: The rapid expansion of large-scale models and datacenter infrastructure—including more than $16 billion in new investments from major technology firms across Europe—places unprecedented pressure on electricity systems (Wall Street Journal, 2025). Rising demand slows the shift to renewables, leading to renewed reliance on natural gas and nuclear power to secure a 24/7 supply (Luscombe, 2024).
Geopolitical and Material Externalisation: AI requires intensive water use for cooling, continuous mineral extraction, and frequent hardware turnover. These practices externalise pollution and e-waste to less-regulated regions, reinforcing global inequalities and turning AI into a mechanism of resource appropriation (Crawford, 2021).
Policy Proposals and Call to Action
Open the Black Box and Mandate Full Lifecycle Transparency: Current governance focuses on algorithms and outputs, leaving the material infrastructures that sustain AI, from mineral sourcing and manufacturing to model training and hardware end-of-life, largely unexamined. These infrastructures are not neutral but reflect institutional priorities and engineering assumptions. To address this, regulators must adopt a holistic lifecycle approach. This structural commitment should be operationalised by requiring standardised reporting of energy use, water consumption, emissions, and hardware turnover from AI developers and data centre operators. Mandating transparency is essential for accurate environmental accounting, effective oversight, and ensuring that mandatory lifecycle assessments identify environmental risks before deployment, not after.
Deploy AI Only Where Necessary: Given the high energy and water demands of large-scale models, AI should be used only where it provides a clear environmental or social benefit. This requires regulatory criteria, currently absent from frameworks like the EU AI Act, that evaluate whether an AI system is truly necessary given its energy and resource footprint. Regulators should promote computational sufficiency and prioritise lower-impact approaches when they achieve comparable results.
Align Public Incentives with Climate Goals: Public subsidies, tax benefits, and access to strategic datasets should depend on measurable contributions to emissions reductions, renewable-energy integration, or improved resource efficiency.
Strengthen Environmental Standards for AI Infrastructure: Datacenters must meet strict efficiency requirements, operate on verifiable renewable energy, and incorporate advanced cooling and heat-recovery systems to reduce pressure on energy and water resources.
Regulate Extraction and E-Waste Across the Supply Chain: AI’s dependence on mineral extraction and rapid hardware turnover requires AI-specific extensions of existing Extended Producer Responsibility (EPR) rules, including binding standards for responsible sourcing, strict limits on e-waste exports, and requirements for repair, reuse and recycling. Such measures would prevent environmental harms from being shifted to less-regulated regions.
References
Brevini, B. (2020). Black boxes, not green: Mythologizing artificial intelligence and omitting the environment. Big Data & Society, 7(2), 2053951720935141.
Crawford, K. (2021). Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence. Yale University Press.
Hao, K. (2021). Training a single AI model can emit as much carbon as five cars in their lifetimes. MIT Technology Review.
Johnson, M., Pereira, L., & Ramaswamy, A. (2025). Artificial intelligence for sustainability: A systematic review and critical analysis of applications and challenges. Sustainability, 17(17), 8049.
Luscombe, R. (2024). Three Mile Island Nuclear Reactor to Restart to Power Microsoft AI Operations. The Guardian, September 20.
Rohde, F., Wagner, J., Meyer, A., Reinhard, P., Voss, M., Petschow, U., & Mollen, A. (2024). Broadening the perspective for sustainable artificial intelligence: Sustainability criteria and indicators for Artificial Intelligence systems. Current Opinion in Environmental Sustainability, 66, 101411.
Vinuesa, R. et al. (2020). The role of artificial intelligence in achieving the Sustainable Development Goals. Nature Communications, 11, 233.
Wall Street Journal. (2025, November 13). Microsoft to Invest Over $10 Billion to Expand AI Infrastructure in Portugal. Wall Street Journal.