In partnership with
Dear {{ first_name | reader }},
In this week's edition we're looking at why data centres have become the physical face of an industry that still can't explain itself — and what that means for anyone doing crisis or risk work near it.
Dario Amodei ( $ANTHZZX ) spent part of last week defending himself on X, after investor Gavin Baker accused him of talking AI into a corner: warnings about the technology's dangers, Baker argued, have helped manufacture the backlash now stalling data centres across the United States.
Amodei pushed back — the problem, he said, isn't his tone, but a crisis of trust that predates him and will outlast whichever AI executive happens to be doing the warning this year.
I don't think either side of that argument is the interesting one.
AI has an explanation problem: most people who oppose it, support it, or shrug at it couldn't tell you with any precision what it does, how it works, or who answers for it when something goes wrong.
That's unusual for a technology this consequential, and it leaves a vacuum — publics don't tolerate vacuums well, so they fill them with the nearest concrete thing. For AI, the nearest concrete thing has a fence line, a substation, and a noise permit.
I spent ten years at IBM, and it put me inside data centres more often than most people in communications ever get to go. Racks, cabling, hums — about as physical and unglamorous as this industry gets.
Ask most people what "AI" is, though, and you won't hear any of that; you'll hear something closer to a mind, something that thinks, decides, maybe even wants things.
I've stood inside the machinery, and I still notice that gap between what it actually is and what people believe it is. Nobody in this industry has closed it, or seriously tried to — which is the real story worth walking through this week, not the Amodei-Baker spat.
Happy reading.
WAG THE DOG NEWSLETTER | ISSUE WEEK 34, 2026
KEY TAKEAWAYS
AI's core communication failure has nothing to do with tone. The public anxiety around the technology is driven far more by not understanding what it is or who is accountable for it than by the specific words any executive chooses to describe its risks.
Data centres are the one part of AI people can actually point at. Everything else about the technology is abstract — a data centre is a building, a noise, a power draw, and a line on someone's utility bill, which makes it the default target for a distrust that otherwise has nowhere to land.
Every metric on the opposition side is moving in the same direction, at once. Roughly $130 billion in US data centre projects were blocked or delayed in the first quarter of 2026 alone, the number of active opposition groups more than doubled in three months to 833 across 49 states, and Gallup's first national poll found 71 percent of Americans oppose having one built near their home.
Nuclear power ran this exact experiment decades ago. An invisible risk with no physical object took the better part of a generation to organise against; a visible one with a fence line and a permit hearing does it in months.
Pledges don't close a trust gap. Binding commitments might. The projects surviving contact with organised opposition are the ones that changed their physical or contractual reality — enforceable tariffs, disclosed water and load figures, benefit agreements with real penalties — not the ones that changed their press release.