AI's hidden footprint: Ambitions must be matched by resource planning

A recent report by the United Nations University Institute for Water, Environment and Health argues that artificial intelligence (AI) must increasingly be viewed as physical infrastructure, with real demands on electricity, water and land. This is a clear departure from how AI is usually framed in terms of productivity, innovation, jobs, misinformation or privacy. The numbers are striking. Global data centres last year consumed an estimated 448 terawatt-hours (TWh) of electricity, more than the annual power consumption of Saudi Arabia. If the current trends continue, this could rise to 945 TWh by 2030. AI workloads alone already account for around one-fifth of data-centre electricity demand and are expected to claim a much larger share in the years ahead. More importantly, it is a warning against the comforting assumption that technological progress will automatically solve the problem. AI models are becoming more efficient, requiring less energy per task. But as costs fall and capabilities improve, AI is being deployed across more applications, devices and services. The result is a classic example of the Jevons Paradox: Gains in efficiency lower costs, which in turn drives higher overall consumption. 

The implications are already beginning to influence public policy. Monterey Park in California recently voted through a ballot measure to become the first municipality in the United States (US) to permanently prohibit new data centres. Unlike mature economies where concerns about resource consumption are emerging after years of digital expansion, India is still building its AI ecosystem. Meta recently announced an agreement with Reliance Industries on a 168-Mw AI-enabled data centre in Jamnagar, Gujarat, alongside commitments on nearly 1 Gw of renewable-energy capacity. Through the IndiaAI Mission, semiconductor incentives and state-level policies, the country is actively encouraging investment in data centres and AI infrastructure. This is not merely about attracting investment. Only 16 per cent of countries host AI-specialised cloud infrastructure and most of the global capacity is concentrated in the US and China. Building domestic AI infrastructure is, therefore, increasingly tied to technological sovereignty and strategic autonomy. 

The challenge is that many of the states attracting these investments are already grappling with water stress and rising electricity demand. Maharashtra, Tamil Nadu, Karnataka and Telangana are among the leading destinations for data-centre investment. They are also regions where water availability and power infrastructure are under increasing pressure from urbanisation, industrialisation and climate change. Clearly, this does not mean India should slow its AI pursuits. The potential of AI in health care, agriculture, energy management and climate resilience is immense. But it does suggest that policymakers should begin treating AI infrastructure as they would any other major infrastructure sector. Environmental disclosures, water-use accounting, energy planning and local resource assessments should become routine. Policymakers must also prepare for the growing challenge of AI-linked electronic waste, which the report estimates could reach 2.5 million tonnes annually by 2030. As the race to build AI infrastructure unfolds, deepening pressures on water, electricity and land will need to be handled carefully.