The Environmental Cost of AI Nobody Wants to Talk About
When Microsoft announced in 2024 that its carbon emissions had increased by 30% since 2020, the culprit was clear: artificial intelligence. The company's aggressive AI investments: data centers, training runs, inference at scale: had blown past its own climate commitments. Google reported similar numbers. So did Amazon. The tech industry had spent years positioning itself as a climate leader. AI was quietly making that impossible.
Training a single large language model like GPT-4 is estimated to consume as much electricity as 120 U.S. homes use in a year. But training is a one-time cost. The real energy drain is inference: the millions of queries processed every day. A single ChatGPT query uses roughly 10 times the energy of a Google search. As AI becomes embedded in everything from email to search to coding tools, the cumulative energy demand is staggering.
The International Energy Agency estimates that data center electricity consumption will double between 2024 and 2026, driven primarily by AI. In some regions, new data center construction is straining electrical grids that were never designed for this kind of demand.
Energy gets the headlines, but water is the quieter crisis. Data centers need water for cooling: lots of it. A single data center can consume millions of gallons per day. In water-stressed regions like Arizona, Texas, and parts of the Middle East, AI data centers are competing with agriculture and residential use for a finite resource.
Google disclosed that its water consumption increased 20% in 2023, largely due to AI workloads. Microsoft's increased by 34%. These numbers will only grow as more data centers come online.
The AI industry's defense is twofold. First, hardware is getting more efficient. Each generation of chips does more computation per watt. Second, AI itself can reduce emissions in other sectors: optimizing energy grids, improving building efficiency, accelerating materials science for better batteries and solar cells.
Please enable JavaScript to read the full article.