A developer's daughter sparked a week-long investigation into AI's environmental footprint. A single ChatGPT query uses about 0.3 watt-hours of electricity—10 times more than a Google search—and requires roughly 500ml of water for data center cooling. With billions of daily queries, the cumulative impact is massive, consuming gallons of water per prompt and millions daily across data centers.

The economic promise of data centers is often overstated. The entire U.S. has only 23,000 permanent data center jobs, a tiny fraction of national employment. OpenAI's Stargate project in Texas, for instance, will create just 100 permanent jobs after construction, subsidized at an average of $1.95 million each. In Virginia, auditors found the state generates only 48 cents in economic benefit for every dollar of tax incentive offered.

Developers can significantly reduce this footprint through smarter choices. Selecting a smaller model, like an 8B parameter version, uses 60 times less energy than a 405B model. Prompt engineering can slash token usage by 30-50%, and caching systems can cut costs and energy by 90% for repeated prompts. Simply starting new conversations instead of continuing long threads also saves compute.