Actionable insights — The AI Boom Is Starting To Crack

4:37 1. Watch AI capex costs crossing into consumer prices — the cycle's first crack

1. Track where the boom's costs land: as long as AI spend is funded by investors and bought voluntarily (paid AI subscriptions), the cycle is self-contained and hard to time.
2. Flag the regime change when those input costs become involuntary — embedded in the price of mainstream physical goods consumers must buy, not opt-in software.
3. Treat the first such pass-through as a leading warning: it converts a corporate-budget story into a consumer-inflation story that can break demand and end the cycle.

• Mainstream-product price hikes attributed to AI/component costs; management blaming a supplier; a Mag-7 selloff on the news (AMZN/MSFT/META all fell) — the floodgate signal.

1:29 2. Decompose a blowout into price vs volume

1. When a cyclical posts eye-popping growth, don't stop at the headline — split the revenue/earnings gain into price (same units, higher ASPs) vs volume/innovation (more units, new products).
2. A gain that is overwhelmingly price is a cyclical pricing spike, not durable unit demand — strong for the seller now, but fragile and prone to mean-reversion.
3. Use that read two ways: stay cautious on owning the price-spike beneficiary, and follow the price increase downstream to see who has to absorb or pass it on.

• Record margins on a flat/old product line; pricing power concentrated against a few large customers; the customer publicly objecting to the price.

15:02 3. Separate a flow-driven drawdown from a broken business — use a control case

1. When a quality name keeps falling, resist the urge to "plug in some material fundamental reason" — first test whether the move is company-specific at all.
2. Find a control: an unrelated company with different fundamentals but similar style/flow exposure, and overlay the price charts.
3. If the two charts are near-identical, the cause isn't either company's fundamentals — it's sector rotation / fund flows. Fade the invented narrative and judge the business on its own numbers.

• Two unrelated quality names dropping in lockstep; "nobody wants to own these companies" sentiment; money visibly rotating into the crowded theme.

8:39 4. Steelman each bear case, then knock it down individually

1. Collect the bear thesis as discrete claims (here three: desperate for M&A, no hits, decelerating growth) rather than a vague "it's broken."
2. Address each on its merits — is it false (test against the company's actual business model), true-but-immaterial, or true-and-priced?
3. Concede the valid ones (he grants the "no recent hit" point) so the surviving view is honest, then weigh what's left against the price.

• A bundled bearish narrative; claims that contradict the company's own business model; "true but already in the multiple" objections.

13:03 5. The concentration test for hit-driven content/platform businesses

1. For a content or creator platform, ask what share of engagement/revenue any single hit, show, or creator represents.
2. If no single item exceeds a low threshold (here ~1% of watch time), the business is a diversified aggregator — durable to any one title cooling off.
3. Re-rate "they haven't had a hit lately" from a thesis-breaker to a non-event for that kind of business.

• A platform where the top title is a tiny share of usage; breadth of catalog/creators; vs concentrated peers (Paramount/HBO) that live or die by a few shows.

19:56 6. Discount media narratives — and check the source's incentive

1. When negative coverage clusters on a name, ask whether the outlets have a structural conflict with the company (a competitor for ads/attention, or a target of its product).
2. Back-test the outlet's track record on the same name — pull the historical headlines and see how the confident bearish calls aged.
3. Weight durable fundamentals (growth, moat, valuation, optionality) over narrative, and use the press-driven drawdown to add.

• An outlet with an incentive against the company; a long history of premature obituaries; fundamentals diverging from the narrative.

18:40 7. The "judgment calls" rule — separate disliking an action from selling the stock

1. When a holding does something you object to (a product, a policy), don't reflexively sell — distinguish a moral/taste objection from a thesis-breaking change to the economics or moat.
2. Accept that almost every company does something you'd do differently; reserve selling for genuine deterioration, not disapproval.
3. Keep a mental benchmark of a name you like almost wholesale to calibrate how much objection is "normal."

• An emotional sell impulse driven by a single disliked action; whether the objection actually impairs growth/moat/cash flow — if not, it's a judgment call, not a sell.