Actionable insights — Should you buy Microsoft after the decline?

1. Split maintenance capex from growth capex before judging capital intensity

1. Compute the raw ratios first — CAPEX/Sales and CAPEX/Operating Cash Flow — and note whether they fail your threshold.
2. Ask whether the spending replaces existing assets (maintenance) or builds new capacity (growth). Only the first is a true cost of staying in business.
3. Use depreciation and amortisation as the proxy for maintenance capex, and treat the excess as growth investment.
4. Recompute the ratios on maintenance alone, and state both sets of numbers so the adjustment is visible rather than assumed.

• Growth capex that turns out to be maintenance in disguise. AI accelerators depreciate on a much shorter cycle than buildings, so today's growth spending can become tomorrow's replacement cycle.

2. Distinguish a declining ROIC caused by investment from one caused by decay

1. When return on invested capital falls, check the denominator before the numerator: is capital rising faster than profit, or is profit falling?
2. Rising capital with stable profit points to a build-out; falling profit points to competitive erosion.
3. Require a stated timeline for the investment to earn its return, and record it so the claim can be tested later.
4. Treat the recovery as a hypothesis, not a fact — this is the same claim every capital-cycle disappointment starts with.

• The two arguments leaning on each other. If the AI capex does not earn a return, both the ROIC explanation and the maintenance-capex adjustment fail together.

3. Charge stock-based compensation in the valuation, not just in the commentary

1. Measure SBC as a percentage of net income, on both the latest year and a five-year average.
2. Apply a hard threshold (here: preferably under 4%, and certainly under 10%) and record a fail as a fail even for a great business.
3. Then actually deduct the dollar amount from free cash flow in every valuation model you run.
4. Never net it off against buybacks without saying so — buybacks that only offset dilution are a cost, not a return.

• An add-back on one line and a deduction on another. Here $12.3bn of SBC is subtracted while $40.9bn of growth capex is added back, and the net effect flatters the base.

4. Interrogate the base of a reverse DCF, not just its answer

1. Write down the free cash flow starting point explicitly, with every adjustment listed as a separate line.
2. Solve for the growth rate the current price requires to deliver your target return.
3. Compare that required rate with your own expectation, and state whether the gap is comfortable, tight or negative.
4. Then re-run it on the unadjusted cash flow, because the adjustments — not the growth rate — are usually where the conclusion is decided.

• Growth capex added back for a business that must keep spending it. The add-back is only honest if the spending is genuinely optional.

5. Let a split verdict across the screens stand as the answer

1. Record each valuation screen's result separately: pass, fail, or uncertain.
2. Do not average them into a single number — a two-pass-one-uncertain result is a different statement from a three-pass result.
3. Where the uncertain screen is the cash-based one, weight it most: it is the test least dependent on the multiple re-rating.
4. Then say what the split means for action, which is the step this article skips.

• A composite score standing in for a decision. 7.9/10 is a description, not an instruction, and the article ends there.

6. Read mandated share ownership differently from granted ownership

1. Separate what an executive was given from what they are required to hold — the second is a constraint, the first is compensation.
2. Express the requirement as a multiple of base salary, which makes it comparable across companies.
3. For a very large company, accept that total insider ownership will be immaterial as control (here under 1%) and judge alignment on the requirement instead.
4. Cross-check with employee sentiment as a secondary signal on culture.

• The requirement being satisfied entirely by unvested grants, which converts a constraint back into compensation.

7. Price an embedded stake as part of what you are buying

1. List the material minority stakes the company holds and their percentage.
2. Decide whether the stake is an asset (value you get for free), a driver (it shapes the operating business), or a risk (it commits capital and reputation).
3. Check whether the same stake appears on both sides of your own analysis — as upside and as risk — and resolve it rather than counting it twice.

• Stakes with no marked value at all. If it is material enough to mention, it is material enough to size.