AI uses real resources. Training and serving models require chips, electricity, data centers, cooling, networking and maintenance. Water consumption and local energy pressure are legitimate concerns. Blaming 'AI' as though it designed its own cooling system, however, obscures the decisions that actually determine the footprint.
Follow the chain
Operators choose facilities, cooling systems and utilization patterns. Power systems determine generation impacts. Chipmakers shape efficiency and supply-chain demand. Customers and application developers influence workloads. Regulators and communities shape the boundaries within which these choices occur. Responsibility must be traced across the mechanism.
What the evidence says
The International Energy Agency reported that data centers consumed around 1.5 percent of global electricity in 2024, approximately 415 terawatt-hours. That is an aggregate figure for data centers, not an attribution of all data-center demand to AI. Local effects vary significantly with region and infrastructure.
A fair evaluation
Measure water withdrawals and consumption separately, where data are available. Examine local water stress, the power mix, the cooling technology and the useful outcome produced. Compare alternatives rather than treating a single headline number as the whole lifecycle assessment.
The October 15 insight
John Brajer challenged the habit of making artificial intelligence the scapegoat for resource decisions taken by human institutions. The strongest version of that argument does not deny environmental costs. It identifies the people and systems with the power to improve them.
Good criticism should tell us what to redesign, not simply what to blame.