The market obsesses over who has the best chips. The more interesting question in mid-2026 is who has the electricity to run them — and this week the power side of the AI trade flashed several red lights at once.
The numbers are getting silly
Google’s environmental report revealed a record 37% year-over-year jump in electricity consumption — its largest ever — driven by AI data centers that alone burned over 42 million megawatt-hours. That’s in the neighborhood of the annual electricity use of an entire country like New Zealand or Denmark. From one company’s data centers. The AI buildout is now outpacing grid decarbonization, and the physics doesn’t care about your net-zero pledge.
And communities are pushing back
This is where the trade gets real. Blackstone-owned QTS just killed a major Virginia data center project after years of local opposition. Pennsylvania voted to pull a $517M big-tech tax break. Meta had data-center water discharges suspended after contaminating a city’s reclamation supply. The bottleneck on AI is quietly shifting from “can we get chips” to “will the town let us plug in and turn on the hoses.”
The AI tool stack actually worth paying for
One email a week. The models, tools and moves that matter, stripped of hype and filtered so you don’t have to drink from the firehose. Free, and you can bail anytime.
Follow the money, not the models
Watch how the capital is repositioning. Nvidia is now offering AI clouds compute now, pay later — revenue-sharing and credit structures instead of huge upfront chip buys — because the constraint isn’t demand for GPUs, it’s the ability to finance and power them. The picks-and-shovels layer of this boom isn’t only silicon. It’s transmission lines, substations, gas turbines, nuclear restarts, cooling, water rights, and the political capital to build in someone’s county.
The market-watcher’s read
If you’re mapping the AI trade, don’t stop at the chipmakers. The names quietly gated by this dynamic are the utilities, independent power producers, grid-equipment suppliers, and data-center REITs — plus the local politics that can cancel a billion-dollar project with one zoning vote. AI’s growth curve is now a power curve, and power is slow, physical, and contested. That’s the part the model-benchmark headlines keep missing.
Not investment advice — a market watcher’s notes only.