2:51 1. The sector-weight concentration screen — chart one theme's share of the index over time
The repeatable method
- Pick the theme leading the market and plot its weight (percentage of the S&P 500), not its price, going back decades — so you're measuring how much of the index one group has become.
- Anchor today's reading against its own history and prior manias: what was the long-run normal, what was the last bubble peak, and how fast did it get here?
- Read the mirror image: as one theme's weight rises, "everything else" falls by the same amount — that de-rating majority is your candidate pool.
Here: semiconductors went ~1% (1995) → ~8% at the 2000 peak → mostly ~2% and never above ~5% for two decades → 8% (2024) → 19.7% in one year, ~20% of the S&P today; "everything else" fell from ~98% to ~80% of the index weight — "unprecedented."
Watch for
- A single sector's index weight going vertical past its prior bubble peak; the complementary "everything else" bucket shrinking fast — the tell that money is reallocating indiscriminately, not pricing fundamentals.
8:32 2. Treat the short-term market as a reallocation game — hunt what's being sold to fund the crowd
The repeatable method
- Frame the near term explicitly: "the market in the short term is mostly a reallocation game" — for one theme to gain that much weight, money has to be pulled from everywhere else quickly.
- Assume the selling of the non-theme names is mechanical (funding the crowded trade), not a verdict on their businesses — so quality gets thrown out with the rest.
- Screen the orphaned bucket for the highest-quality, most durable compounders now at multi-year-low valuations — "a fertile hunting ground."
Here: non-semi big tech "at multi-year lows" despite accelerating revenue and stable headcount; he pulled nine names out of the orphan bucket (META, AMZN, MSFT, NFLX, UBER, DASH, CPRT, CNSWF, MA) as "left-behind quality."
Watch for
- Great businesses de-rating on no company-specific bad news; a chart that moves in lockstep with an unrelated peer (here Netflix and Spotify) — proof the move is flow, not fundamentals.
6:42 3. Use the 2000 analogy — the alternative to selling is buying outside the bubble
The repeatable method
- When you fear the hyped leaders are overvalued, resist the binary of "stay fully in the mania" vs "get out of the market" — there's a third option.
- Look at what the last comparable bubble taught: the danger was concentrated in the crowded names; the rest of the market was often a great buy from the very peak.
- Rotate toward the un-hyped survivors — the businesses "the only place you didn't want to be was the place everyone wanted to be."
Here: from the 2000 top, non-internet stocks did well — consumer staples +35%, plus utilities, healthcare, small caps, railroads, Monster Energy — while the dot-com names were the specific ones to avoid. The parallel: buy the compounders being orphaned by the semi trade.
Watch for
- A mania narrow enough that quality outside it is cheap; historical precedent that the crash was contained to the crowded theme, not the whole market.
11:09 4. The EBIT-per-employee operating-leverage screen
The repeatable method
- For a company investing heavily in AI, test whether the spend is producing operating leverage: track adjusted EBIT per employee across years.
- The bullish signature is profit-per-head rising while headcount stays flat or falls — evidence the technology is doing work that used to require more people.
- Compare the CAGR of EBIT/employee across peers to rank who is compounding operating leverage fastest.
Here: GOOGL $393k (2022) → 488 → 613 → 676, rising through 2028; MSFT $389k → past $1M; META toward ~$1.78M at a faster CAGR (headcount falling); AMZN $8k → $112k (2028), fastest CAGR of all — the quantitative backbone of the big-tech buys.
Watch for
- Rising EBIT/employee with flat-or-shrinking headcount and accelerating revenue; a widening gap between profit growth and hiring — real AI operating leverage, not just a spending story.
16:37 5. Value a quality name against its own valuation history, not the market's
The repeatable method
- For a durable compounder, chart its trailing PE over ~5–10 years to establish its normal band and its prior stress-lows.
- Flag it when today's multiple sits at the bottom of that personal range — especially if it matches a known crisis low while the business is intact.
- Cross-check the fundamentals still hold (growth intact, share not lost) so you're buying a de-rating, not a deterioration.
Here: MSFT normally trades 30–38× (near 40× in 2024) but sits at a 5-year-low ~23× — "the exact 2022-selloff bottom" — with high-teens EPS growth intact and Office not losing share. MA at ~23× 2027 earnings and NFLX below 20× are the same test.
Watch for
- A high-quality name at the floor of its own multiple range that coincides with a past panic low; unchanged growth/market-share to confirm the cheapness is flow-driven.
27:20 6. Separate real AI-disruption risk from indiscriminate "software distaste"
The repeatable method
- When a whole category sells off on an AI-will-replace-it narrative, don't accept it wholesale — ask whether AI actually threatens this specific product.
- Judge the stickiness: mission-critical, niche, deeply embedded software is far harder for a general model to rebuild than a single-feature app.
- If the fear is generic ("distaste for anything software") rather than specific, treat the sell-off as overdone and the moat as intact.
Here: CNSWF Constellation Software sold on general software fear; Carlson: "I do not believe that Claude is going to replace… the huge majority of software that they own… selloff is likely overdone." Same logic keeps CPRT's real-estate/insurance/software moat intact.
Watch for
- A whole sector de-rated on a blanket AI-disruption story; individual businesses whose product is niche/embedded enough that the threat doesn't actually apply.
31:51 7. Avoid the premium-to-NAV roll-up flywheel — it runs in reverse
The repeatable method
- Identify the structure: a company that issues stock to buy an asset, and whose whole model depends on trading at a premium to the value of what it holds (so each raise is accretive).
- Recognize the fragility: the flywheel only spins while the premium exists. Watch the premium/discount to NAV as the single key metric.
- Assume asymmetry: once it flips to a discount, the same mechanism reverses and unwinds — "roll-ups typically don't work well in reverse" — and leverage amplifies the fall.
Here: MSTR down ~80% ($450+ → ~$98) — a levered Bitcoin play whose premium-to-NAV model unraveled the moment it began trading at a discount; Carlson calls the collapse "entirely inevitable."
Watch for
- A vehicle valued above its net asset value that relies on issuing shares to buy more of the same asset; a narrowing premium heading toward a discount — the trigger for the reverse flywheel.