# The Math Behind AI's Environmental Impact Source: https://humanspark.ai/ai-and-energy/ 4,500x how wrong a bestselling book's water claim was 30x more water used by US golf courses than data centres 0.26mL water per Gemini query - about five drops 1.5% data centres' share of global electricity (2024) 662% gap between big tech's location- vs market-based emissions I've been increasingly frustrated watching the environmental conversation around AI collapse into bumper stickers - "AI is boiling the oceans" on one side, "everything's fine" on the other. Neither is useful. Both are wrong. > If your opinion on AI can fit on a bumper sticker, you're probably wrong. The bad data floating around makes the nuanced conversation we actually need nearly impossible. So I dug into the most-shared claims. Here's one that captures the problem perfectly. ## 01The 4,500x error A bestselling book claimed a single Google data centre in Chile would use "more than one thousand times the water consumed by the entire population of Cerrillos - roughly eighty-eight thousand residents." One building, using 1,000x the water of 88,000 people - i.e. the same as a city of 88 million. That's Tokyo. Plus Paris. Plus London, New York, Sydney, Rome, Los Angeles, Dublin, *and* the rest of Ireland. Published by Penguin Random House. Author with an MIT engineering degree. Praised by the New York Times and The Economist. A fact-checking team thanked in the acknowledgements. The claim is wrong by a factor of around 4,500. Data scientist Andy Masley did the maths: the figure implies each resident uses 0.2 litres a day - a fifth of a water bottle, when adults need 2-4 litres just to stay alive. // what actually happened The source reported water usage in cubic metres. The book recorded it in litres - a 1,000x unit error, built into the central claim, sailing through fact-checkers, editors and reviewers at every major publication. Nearly 1,000 Amazon reviews, but a solo blogger caught it first. ## 02What the water numbers actually show Google says a median Gemini query uses 0.26 millilitres of water - about five drops, not the "half a bottle" of viral claims. In aggregate, US data centres consumed ~17.4 billion gallons in 2023; that same year US golf courses used 548 billion gallons. **Golf courses use 30 times more water than data centres.** An important caveat: indirect water use (at the power plants generating the electricity) is far larger - about 800 billion litres in 2023 - and the energy source matters enormously. The impact is real and worth taking seriously, but it's nowhere near "draining the aquifers" - and exaggerating it thousands-fold makes the actual problem harder to address. ## 03The energy picture Global data centres used about 415 TWh of electricity in 2024 - roughly 1.5% of global electricity and 0.9% of energy-related emissions. Real impact, worth attention, nowhere near apocalyptic. > Did you reboil the kettle this morning? Drive somewhere you could have walked? Any one of those uses more energy than hundreds of AI queries. Efficiency varies enormously by facility. Power Usage Effectiveness (PUE) measures overhead - cooling and infrastructure, everything that isn't computing. Lower is better. | Facility type | Typical PUE | | --- | --- | | Older enterprise / on-premise | 1.5 - 2.0 | | Industry average | 1.55 - 1.8 | | Hyperscale (Google, Microsoft, AWS) | 1.09 - 1.2 | Google's fleet-wide average is 1.09 - 84% less overhead than the industry average. Practically: running AI on ageing on-premise servers? Moving to hyperscale cloud probably reduces your carbon footprint per unit of computation. That's genuinely good news. Here's what complicates it. ## 04The growth problem The IEA projects global data-centre electricity demand to double by 2030. In the US, Lawrence Berkeley National Laboratory estimates 176 TWh in 2023 rising to 325-580 TWh by 2028 - potentially tripling in five years. The driver is AI: GPU workloads grow ~30% a year versus 9% for conventional servers. Individual queries get more efficient, but we run vastly more of them. > The environmental cost per query is dropping. The total environmental cost is rising. Both are true - and neither fits on a bumper sticker. ## 05The accounting gets murky When Google says it "matches 100% of its electricity with renewable energy," that's technically true and somewhat misleading. It buys Renewable Energy Certificates equivalent to its consumption - but those might come from a Texas wind farm while a Virginia data centre draws from a coal-heavy grid. The maths balances on paper; the electrons don't match. A 2024 Guardian investigation found location-based emissions for Google, Microsoft, Meta and Apple were 662% higher than their market-based figures. To their credit, Google and Microsoft now pursue hour-by-hour "24/7 carbon-free energy" - harder, and more honest. If you're evaluating a provider, ask which accounting method they use. ## 06What this means for you Hyperscale cloud probably reduces your per-computation footprint. Google at 1.09 vs your server room at 1.7 is a meaningful difference. "Cloud is green" isn't a free pass. Efficiency gains are real, but so is demand growth - more efficient infrastructure doing vastly more work doesn't automatically shrink your footprint. Water claims need sanity-checking. If someone cites an alarming statistic, do the maths - the real numbers are orders of magnitude smaller than viral claims. Emissions accounting is complicated. RECs and hour-by-hour matching are different things; know which your provider uses. Question everything - including this article. It's referenced to the IEA and Lawrence Berkeley National Laboratory; check the sources yourself. I'd rather deal with the real trade-offs than pick a team and defend a bumper sticker. Hyperscale efficiency is real. Demand growth is real. The honest position sits in the middle - less shareable, far more useful. [Book a Focus Call](https://humanspark.ai/call/)