Toolkit · Worked examples
Each name ships in two flavors so methodology disagreement is explicit. Through-cycle uses winsorized 5y mid-cycle margin (drops the trough outlier); AI/expansion-cycle uses peak-half average (assumes the current expansion is structurally different). The spread between them IS the recommendation.
The bear/skeptic case. 5y mid-cycle EBITDA margin computed by dropping the FY23 trough AND the FY21 peak before averaging. Treats the AI demand ramp as another up-cycle that mean-reverts. Assumes 8x exit multiple. Ships with reverse-DCF showing what perpetual growth rate today's price requires.
The structural-shift case. 5y mid-cycle margin uses ONLY the upper half of years (assumes HBM long-term agreements put a floor under memory margins that didn't exist in prior cycles). Faster Y1-Y3 revenue ramp, higher capex intensity (HBM3/HBM4 capacity build), 11x premium entry multiple for the AI exposure. WARNING in the workbook: becomes wrong if HBM commoditizes.
| Driver | Through-cycle | AI cycle | Why |
|---|---|---|---|
| Mid-cycle method | Winsorize (drop best+worst) | Peak-half (top half only) | Bear assumes mean-reversion; bull assumes HBM puts a floor |
| Y1 revenue growth | 20% | 35% | HBM ramp pulls forward more demand in bull case |
| Capex intensity | 25% | 32% | Bull case requires higher capex to build HBM3/HBM4 capacity |
| LBO entry multiple | 8.0× | 11.0× | Premium for AI optionality |
| Terminal growth | 2.5% | 3.0% | Bull assumes longer secular tailwind |
Forthcoming examples. XOM (energy), NVDA (semis hyperscaler), and a tier-list of integrated oils (XOM/CVX/COP/SHEL/BP) on capital-allocation discipline are queued. Each will ship as paired through-cycle / expansion workbooks.
Want a worked example for a specific ticker? Request the pipeline and I'll generate it. Or run it yourself.