Actionable insights — Oracle: 850 Megawatts Later

1. Test a backlog story on capacity delivered and growth acceleration — not on the size of the backlog

1. When a company's thesis rests on a huge contracted backlog (RPO), treat the backlog as a promise whose delivery is physically constrained — find the unit of delivery (megawatts online, GPUs deployed, stores opened, units shipped).
2. Track that unit quarter by quarter. A backlog converts only as fast as capacity comes online; if capacity isn't stepping up, revenue can't either, whatever the RPO says.
3. Check that the related revenue line accelerates as capacity arrives (growth rate rising, not merely high). Acceleration in step with capacity is the proof that demand was waiting for supply.
4. Then check the backlog still grows sequentially after conversion — new bookings outpacing what's being consumed means the pipeline isn't being drawn down to make the quarter.
5. Note what the argument has changed to: once conversion is visible, the bull case stops being "is the demand real?" and becomes "are the economics worth the capital?" — re-underwrite on the new question.

• Megawatts delivered per quarter flattening while RPO keeps rising (the gap reopens); Q2 FY27 cloud growth landing inside the guided 65%–71%; and RPO growing slower than cloud revenue, which would mean the backlog is being consumed faster than replenished.

2. Net the capital bill against who is actually paying it — reported CapEx overstates the company's own cash at risk when customers prepay

1. Start from reported CapEx and free cash flow, quarterly and trailing twelve months, so one lumpy quarter doesn't hide the trend.
2. Find customer prepayments / deferred revenue collected for the buildout and subtract them: net CapEx = reported CapEx − customer-funded portion. That is the company's own capital at risk.
3. List the remaining funding sources explicitly — operating cash, debt, equity issuance — and whether each is done or still pending.
4. Look for a management statement tying new bookings to funding needs ("no incremental increase to the planned capital raise"). New contracts that don't require new capital are a better signal than new contracts, full stop.
5. Remember the other side: prepayments shift financing risk to the customer, so the customer's own funding capacity becomes part of your diligence.

• The prepayment share of CapEx each quarter (rising = customers carrying more of the build); progress against ~$70B of net CapEx this year; any new raise despite the "no incremental increase" statement; and TTM FCF bottoming as converted revenue catches up.

3. When gross margin falls and operating margin rises in the same quarter, diagnose it as mix — then find what is absorbing it

1. Read gross and operating margin together. Opposite directions mean the story is in the P&L structure, not in pricing or demand.
2. Check segment gross margins: if the fastest-growing segment carries a structurally lower margin, falling consolidated gross margin is arithmetic (mix), not deterioration.
3. Identify what is offsetting it below gross profit — opex growing far slower than revenue (R&D, S&M, G&A as a share of sales). That is operating leverage.
4. Ask how long the offset lasts: operating leverage is finite if opex can't keep shrinking as a share, while mix pressure grows as the low-margin segment compounds.
5. Model the crossover — the point where continued mix shift outruns remaining leverage and operating margin starts to fall too.

• OCI's share of revenue (38% this quarter and rising); operating-margin expansion slowing while gross margin keeps falling; and any OCI-specific gross margin disclosure, which would replace the inference with a number.

4. Value a premium product on the dollars it adds — subtract the purchase the customer would have made anyway

1. Compute the headline: units × average price. Label it gross, not incremental.
2. Ask what the buyer would otherwise have bought from the same company. For a flagship upgrade inside an existing franchise, the answer is usually "the next most expensive model."
3. Incremental revenue = (new price − substitute price) × units, plus any genuinely new customers at full price. Add upsell tiers (storage, accessories) separately.
4. Compare the incremental figure with the company's base in that product line — that ratio, not the gross headline, tells you whether the launch matters.
5. Apply it to any "new product adds $X billion" claim: a cannibalizing launch is a price increase in disguise, and should be analysed as one.

• Duo unit estimates (6M in 2026, ~10M first year) versus Pro Max mix; storage-tier attach on the Duo; and whether Apple ever discloses switchers from Android, which would be the only fully incremental buyers.

5. Separate share of the category from share of the company — you can dominate a market that doesn't move you

1. Size the category as a share of the parent market (foldables as a % of all smartphones).
2. Estimate the company's likely category share, then convert it back into the company's own volume: category units ÷ company total units.
3. If the category is small, a dominant share can still be immaterial — report both numbers side by side.
4. Then ask whether the category can grow fast enough (or whether the entrant will expand it) to change the second ratio within your holding period.
5. Use the lateness of an entry as data, not a flaw: an incumbent entering a years-old, still-tiny category is betting on its distribution, not on the category's proven size.

• Foldable share of total smartphone shipments rising after Apple's entry (the category expanding rather than Apple just taking share); Samsung foldable volumes; and Duo units approaching 5% of iPhone volume.

6. When hardware growth relies on price, read the lineup — removing the cheap option is a price increase

1. Split product revenue into units and ASP. If units are flat, all growth is price — note it explicitly.
2. Track list-price changes across the whole line, including older models, and attribute them (input costs such as memory versus chosen premiumization).
3. Check what disappeared: dropping the entry model of a new generation forces anyone who wants the latest into a higher tier — a mix-driven ASP increase that never appears as a "price hike."
4. Test whether the price increase is sustainable: pass-through of an input cost can reverse when the input normalizes; lineup-driven trade-up tends to stick.
5. Compare with the retailer or supplier side of the same input cost to see where the price is actually landing.

• iPhone unit commentary in the next print; whether a standard model returns in spring; memory prices normalizing (does the $100 stay?); and Services growth, the non-hardware engine that doesn't depend on price increases.

7. Classify each company's AI monetization as direct or indirect — and value the indirect kind through the product it protects

1. For each AI launch, ask where the money appears: a new line (subscription tier, usage fee, API revenue) or an existing one (hardware upgrades, retention, attachment, ad pricing).
2. Direct monetization is easy to measure but must win on the AI product itself. Indirect monetization avoids that contest but needs a control point — device, OS, distribution, personal data.
3. For indirect players, don't look for AI revenue; look for the metric the AI is supposed to move (upgrade rate, ASP, ecosystem attach, churn).
4. Check model dependence: a control-point owner that can swap in outside models is insulated from the foundation-model race; one that must build the best model is not.
5. Put rival strategies side by side — the same week can show a platform launching paid tiers and another explicitly refusing to.

• Apple upgrade-cycle and Watch/AirPods attach commentary after the new Siri ships; Muse paid-tier adoption if Meta discloses it; and any Apple move toward a paid AI tier, which would signal the indirect model isn't enough.