A single query processed by a state-of-the-art Large Language Model (LLM) consumes up to ten times more electricity than a standard internet search. Furthermore, training a single foundation model can emit more CO2 than five average cars over their entire lifetimes, whilst cooling the associated high-density graphics processing units (GPUs) requires millions of litres of fresh water. With the rapid integration of GenAI tools into everyday software, corporate workflows, and consumer search engines, the aggregate electricity and water demand from hyperscale datacentres is projected to rise exponentially.
Because major AI developers operate globally and utilize cloud networks that span multiple jurisdictions, unilateral domestic regulation is exceptionally complex. Governments are considering various policy levers, including water-usage levies on cooling infrastructure, minimum computational efficiency mandates for model architectures, subsidizing localized green computing zones, and requiring "water and carbon disclosure labels" on consumer-facing AI interfaces.
Evaluate potential government interventions to address the environmental market failure arising from the rapid growth of generative artificial intelligence services.
