Costco: The Multiple Does the Work A great business is not the same as a great return, and the lens can tell them apart

The finding

Costco's five-year return is decided by the price a future buyer pays, and growth cannot rescue it.

That is not a claim that growth does not matter. It matters a great deal. It is a claim about size: the growth needed to offset even a modest fall in the valuation is beyond anything Costco has ever delivered.

The question the model can answer cleanly is this one — what would the business have to do?

If a buyer in year 5 pays…	…sales must then grow, every year, at
46× cash flow — what buyers pay today	8.6% — reachable; the last five years ran above 10%
40×	11.5% — beyond any five-year stretch in the record
36×	13.7% — beyond any five-year stretch in the record
30×	17.6% — Costco has never grown at this rate
24×	22.6% — impossible

To return 12% a year, with margin drifting up to 4.10% by year 5.

Read the first row and the third together. If the valuation holds exactly where it is, the business needs 8.6% growth, and that is a fair bet: Costco has just done better than that for five years. If the valuation slips to 36× — still a very large premium, not a collapse — the business needs 13.7%, which it has never managed over any five-year period, in any conditions.

So the valuation does not have to fall far. It has to drift, and no achievable operating performance offsets the drift.

This reframes the position. It is not a question about whether Costco is a good business. The data answers that, and the answer is yes. It is a question about whether a great business, bought at a great price by someone else a decade ago, is still a great purchase today.

The business in one read

What it is. Costco sells groceries and household goods out of large warehouse-format stores. You cannot shop there without paying an annual membership fee. That fee is the business.

How it actually makes money. Costco keeps under $4 of operating profit for every $100 of goods it sells — a margin most retailers would consider a failure. That is deliberate. The merchandise is sold close to cost, on purpose, so that the membership stays worth renewing. A large share of the operating profit arrives as membership fees, which cost almost nothing to deliver. The retail operation is not really the profit engine; it is the reason the fee keeps getting paid. (That last point is [ASSUMPTION] — agent-sourced. Segment detail is not in the pack. It explains why the margin line is so steady, and changes none of the arithmetic below.)

What the numbers show. A record of near-mechanical consistency. Sales have grown every single year in the data, without one decline. Operating margin has moved from about 3.1% to about 3.8% over ten years — roughly two-thirds of a point, delivered a basis point at a time, never lurching. Costco collects cash from shoppers before it pays its suppliers, so cash flow reliably runs ahead of reported profit. Returns on the capital it puts into stores have climbed from the mid-teens to the mid-twenties and stayed there. It buys almost nothing: this is growth built store by store, not by acquisition.

Two details cut against the reflexes the rest of this portfolio trains.

The share count goes up, not down. Many mature companies lift per-share results by buying back their own stock, shrinking the number of slices the pie is cut into. Costco barely does this. It repurchases a little each year, and the count still ends up slightly higher than five years ago — so new shares are being issued faster than old ones are retired. Nothing in this note's math should credit buybacks, because there is nothing to credit. (The data shows the count and the repurchases. It does not show what the new shares are issued for — that is not in the pack.)

The dividend line is a decoy. It jumps around by multiples year to year, because Costco periodically pays a large one-off special dividend that lands in the same line. Read as a steady payout it is noise. Any story built on that line is describing the specials, not the policy.

Why this engine

A word on what the tool is, since the sections above lean on it. It is not a forecast. It is a lens: you set a handful of assumptions, and it tells you what return those assumptions imply at today's price. The point is to find out which assumption is actually driving the answer.

The engine chosen is capitalLight — and its name is wrong for this company.

Costco is capital-hungry. It spends roughly forty cents of every dollar of operating cash flow on building and fitting out stores. Nobody would call that asset-light. The fit is not about the label; it is about the math.

capitalLight grows sales, drifts the margin, and turns the result into cash using a conversion ratio fixed at the level the business has actually delivered. That fixed ratio is the point: it holds store-building intensity constant at its real historical level and lets the two things that matter — sales growth and margin — do the moving. Those are exactly the two lines with the longest and cleanest history in the data, and every input the engine needs is present, with nothing missing.

The engine that should have won, and why it didn't. The registry's reinvest engine names Costco explicitly in its own description. It grows a company by asking how much cash it ploughs back and what return it earns on that cash — and on this data, that method reproduces Costco's actual sales growth almost exactly, which is a real point in its favour.

It breaks on its starting input. reinvest values "owner earnings" — roughly, profit minus the spending needed just to stand still. The owner-earnings line for Costco swings by a factor of four across five years, because it subtracts all store spending as if it were merely upkeep, against a company that opens new warehouses continuously. Almost all of that spending is growth, not maintenance. The method is wrong here by construction, and wrong in one direction. Run it anyway and it demands a wildly implausible exit valuation to clear any sensible hurdle — while its own internal scoring caps out far below that. Every outcome it can reach sits outside its own calibration. It cannot produce a supportive reading at any setting.

The routing rule says: where two engines fit, take the one whose math can represent the thesis being false. The corollary earned here is the mirror image. An engine that can only express your thesis will confirm it; an engine that can only express its negation will reject it. Neither is a lens. reinvest can only say no about Costco, and it says no for a reason that is about its own maintenance-spending shortcut rather than about Costco. capitalLight can say either, using two levers we can genuinely argue about.

The honest objection, recorded: the fixed conversion ratio assumes store-building stays as costly as it has been, relative to sales. If growth starts demanding more capital per dollar of sales, cash will come in below the model at every setting and the engine will not warn you. Over the last five years that spending has run between 39% and 53% of operating cash flow — near 40% in four of them, with one year at 53%. Stable enough to lock, not so stable that it can be forgotten. It is the engine's load-bearing assumption and the first thing on the re-check list.

The scenarios

Read the last column as the annual return the assumptions in that row would produce.

	Sales growth	Yr-5 margin	Exit valuation	Yr-5 cash flow	Total vs. today	Annual return
Bull	9.0%	4.60%	46×	$16.3B	2.01×	+14.9%
Base	7.0%	4.10%	36×	$13.3B	1.31×	+5.6%
Bear	4.5%	3.80%	24×	$10.9B	0.77×	−5.1%

The bull case is reachable but demanding, and more demanding than it looks. Sales growing at 9% is fair — below the last five years. The valuation never slipping is a real possibility. The margin is the stretch: 4.60% asks for nearly 0.8 points of drift in five years, when the whole last decade produced roughly 0.65. That is about two and a half times the historical pace, and it is quietly carrying a large part of the bull return.

The bear case is the one to sit with, because nothing in it is a failure. Growth at 4.5% is Costco's own pre-2020 pace, not a collapse. Margin flat rather than falling. The company does nothing wrong. The return is still negative — supplied entirely by the price paid at the start.

All of it recomputes live in the tool, which is where the current price lives.

What the lens is saying

Growth cannot buy back a de-rating. This is the finding and it is worth stating plainly: there is no reachable rate of sales growth that clears a low-teens return if the valuation drifts down even moderately. Not a low rate — no rate. At Costco's best-ever five-year pace, with margin drifting at its true historical rate, a valuation of 36× returns about nine percent and 24× returns under two. The company is not the variable that saves those outcomes, because it cannot be.

A held valuation is the optimistic case, not the neutral one. In most stocks, assuming today's valuation persists is the cautious, boring assumption. Here, persistence is the bull branch — the whole reachable case depends on it. Assuming any drift back toward normal is assuming a weak or negative return.

Beware how far the sliders travel. The exit lever's whole range is plausible; valuations really do move that far. The growth and margin levers reach places Costco has never been — 12% growth exceeds any five-year stretch in its record, and a 5.0% margin is more than three times the historical drift rate. So any impression of "how much each lever moves the answer" flatters the business, because the business levers are being allowed settings the company has never achieved. Judge the levers by what they can reach, not by how far they slide.

The regime panel under-reads this name, and the reason is worth knowing. The engine's internal scoring was calibrated for fast-growing software businesses. Its "fundamentals" score cannot rise past about two-thirds for a company growing at 7%, and its "valuation" score maxes out at a 30× exit — so across the entire upper half of Costco's plausible range, that score is stuck at full marks and the panel goes deaf to the one lever that matters. The band thresholds here are set to the range Costco can actually reach, which keeps them distinct and readable. It does not fix the deafness. Read the return, not the regime, on this note.

What breaks the base case

• Sales growth slowing to its pre-2020 pace. The base assumes 7%, between the last five years and the five before. It is the most defensible assumption in the note and still costs about two and a half points of annual return if it slips.
• Margin drift stalling. Ten years bought about two-thirds of a point. The base extrapolates roughly that rate; the bull asks for two and a half times it. Neither may be renewable — a decade of small wins can be a decade of using up a fixed stock of them.
• The valuation, which is not a thesis but a market fact. No amount of work on this company produces a view on it. That is exactly why it should not be assumed to hold.
• The conversion lock. If store-building gets more capital-hungry per dollar of sales, cash comes in below the model everywhere and the engine stays silent about it.
• The membership dependency, unverifiable from the data here. If fees really are carrying the margin, then margin stability is a fact about renewal rates and fee increases — a different thing to watch than the one the model exposes.

Re-check checklist

1. Store spending as a share of operating cash flow — between 39% and 53% over the last five years, near 40% in four of them. A sustained rise breaks the conversion lock, which is what the whole engine rests on.
   Threshold: two consecutive fiscal years above 50%. One year there has already happened and did not persist.
2. Sales growth against the 7% base. Two consecutive years below 5% moves the base case into the bear's assumptions.
3. Margin trajectory. The base needs roughly six hundredths of a point a year; the bull needs fifteen. Two flat years means the base is extrapolating a finished process.
4. Share count. It rises slowly. If it starts falling meaningfully, the engine choice changes and this note needs re-routing.
5. Valuation history, once the pack ships it. The deciding lever is currently anchored on reasoning alone — the lever that decides the note is the lever with no data behind it.
6. SBC, once schema v15 lands. The cash-flow basis is currently asserted rather than confirmed, and the share-count drift has no disclosed cause in the data.

The one question

Costco is not a question about Costco.

Every hour spent on renewal rates, on Kirkland, on store openings, on the next fee increase lands on two levers: sales growth and margin. Work them as hard as the record allows — Costco's best-ever five-year growth pace, margin drifting at its true historical rate — and a valuation that eases to 36× still returns about nine percent a year. Ease to 24× and it returns under two. The research does not reach the answer, because the answer is set by what the next buyer will pay, and no work on this company produces an edge on that.

This is not a reason to avoid the name. It is a reason to be honest about what the position actually is: not an underwriting of a business, but a wager on the durability of a valuation, dressed in the comfort of an unusually easy business to admire.

So: is there any version of this where work on the business changes the answer? If only the valuation decides, then the entry price is not a detail of the thesis. It is the thesis — and the discipline that applies is the one about waiting, not the one about analysing.