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.
The Energy Numbers
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.
The Water Problem
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 Efficiency Counter-Argument
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.
Both arguments have merit. NVIDIA's Blackwell chips are significantly more energy-efficient than their predecessors. AI-optimized HVAC systems can cut building energy use by 30%. Google's DeepMind reduced the energy needed to cool its own data centers by 40% using AI.
But there's a catch: the Jevons paradox. When something becomes more efficient, people use more of it, not less. More efficient chips mean bigger models, not lower energy consumption. The net effect has been consistently rising demand.
The Nuclear Option
Tech companies have arrived at a surprising conclusion: the only way to power AI sustainably is nuclear energy. Microsoft signed a deal to restart Three Mile Island. Google invested in small modular reactors. Amazon purchased a nuclear-powered data center campus. Sam Altman personally invested billions in Oklo and Helion, nuclear and fusion energy startups.
The logic is straightforward: AI data centers need massive, reliable, carbon-free baseload power. Only nuclear provides that at scale. But nuclear plants take years to build, face regulatory hurdles, and carry their own environmental and safety concerns.
The Accountability Gap
Perhaps the most frustrating aspect is the lack of transparency. AI companies don't disclose the energy cost of individual models. Cloud providers don't break down AI workloads in their sustainability reports. Users have no way to know the environmental cost of their ChatGPT conversation or AI-generated image.
Until there's standardized measurement and disclosure, the environmental cost of AI will remain a number that's easy to ignore. But ignoring it doesn't make it go away. The question isn't whether AI is worth the environmental cost — it's whether we're being honest about what that cost is.