11:04 1. The underspend-reversion screen — buy capex after a long drought
The repeatable method
- Find an industry that has been capex-disciplined for years after repeated boom/busts — "i.e. they just haven't spent."
- Quantify the under-investment with capital-efficiency ratios: profitability-over-capex and revenue-acceleration-over-capex. When both look "anemic" despite strong demand, capacity is too tight.
- Use the leading indicator: "historically how profitable the customers are leads to forward CapEx." Fat customer margins today predict a spending wave tomorrow.
- Own the picks-and-shovels (the equipment makers) ahead of the spend — they get paid first when the chipmakers finally invest.
Here: a decade of discipline with Taiwan Semi the lone (and still under-) spender → buy the equipment makers;
LRCX the memory-levered favorite (
22:42).
Watch for
- Sectors with years of flat capex while end-demand grows; customer gross margins turning up as the forward-capex tell.
12:07 2. Count the spenders — "one → many" breaks the buyer's monopsony
The repeatable method
- Count how many credible buyers of the suppliers' product exist. When there's only one, that buyer "had all the power negotiating with the equipment companies" — bad for supplier pricing.
- Watch for new entrants/returnees that flip it to "multiple spenders, all of which underspent." More competing buyers = the suppliers regain pricing power.
- Confirm with concrete signs of the new spenders showing up (a left-for-dead competitor "picking up customers," fresh foundry announcements).
Here: Intel's foundry "picking up customers" + Samsung announcements → from one spender (Taiwan Semi) to many → equipment pricing power (INTC, SSNLF, TSM).
Watch for
- New or returning capacity buyers in a one-buyer market; foundry/fab announcements that multiply the customer base.
11:43 3. The neglected-cycle radar — "when did we last talk about this?"
The repeatable method
- Ask which cyclical hasn't been a topic in years — "what's the last time we talked about a NAND cycle? Must be a decade ago." Forgotten cycles are under-owned and under-modeled.
- Verify the cycle is real and powerful by the customers' profitability (here memory at ~80% margins — "absolutely enormous").
- Because nobody's positioned, the move can be outsized when the cycle finally turns.
Here: the dormant NAND/DRAM cycle re-igniting → MU, HXSCL, SNDK.
Watch for
- Cyclicals absent from the narrative for years whose end-customer margins are quietly enormous.
12:25 4. The customer-margin pricing-power test
The repeatable method
- Check how profitable the supplier's customers are. When customers run 70-80% gross margins, they're price-insensitive on equipment — so the supplier has pricing power "on top of" unit growth.
- Size the addressable spend against consensus: he sees wafer-fab equipment (WFE) going from a perceived ~$120-130B toward ~$300B over 3-4 years.
- Translate to estimates: where consensus hasn't modeled the inflection, flag the gap ("the estimates are 50 to 70% too low"). A stock that "doesn't screen cheap" can still be cheap on the right numbers.
Here: memory ~80%, Taiwan Semi ~70%, semi-equipment ~50% margins → equipment pricing power; estimates "50 to 70% too low," "next 50 to 100 is up."
Watch for
- Supplier industries whose customers earn fat margins; consensus revenue/EPS that lags a visible capex inflection.
13:12 5. The de-cyclicalization re-rate — LTAs turn a cyclical into a compounder
The repeatable method
- Look for structural changes that reduce a cyclical's volatility — here long-term agreements (LTAs) signed "right and left," giving multi-year demand visibility.
- When a business is "just not as cyclical" anymore, the market should pay a higher multiple — so the re-rating is a second leg of return on top of the earnings beat.
- Add balance-sheet optionality as confirmation: pristine balance sheets that fund M&A and buybacks.
Here: memory makers signing LTAs → less cyclical → potential multiple re-rate for the equipment names on top of the estimate beat.
Watch for
- Long-term supply contracts replacing spot demand; a cyclical sector earning a structurally higher multiple.
1:27 6. Split the AI macro call — inflationary now, deflationary later
The repeatable method
- Resist the single-direction take. Separate the near-term effect (AI's input costs — CPUs, memory, infrastructure — are skyrocketing, and the old economy is hiring to implement it) from the long-term effect (LLMs deliver "a lot more for a lot less").
- Sequence them: "first you get a sticky inflation," then deflation as the technology (and robotics) diffuses.
- Use it for regime positioning — sticky inflation that traps the Fed is bullish the hard-asset / input-cost beneficiaries (the chip-input complex) before the deflationary phase arrives.
Here: skyrocketing CPU/memory input costs + robust labor (+18% software-engineer hiring) = near-term inflation that underwrites the semi-capex trade now.
Watch for
- Input-cost inflation in the buildout phase vs output-price deflation in the diffusion phase; which one the market is pricing.