What money cannot buy quickly.

The circular version

Most moat analysis runs backwards. You observe that a company earns high returns on capital, note that it has done so for a while, and conclude it has a moat. Then you use the moat to justify expecting the high returns to continue.

That is a description wearing the costume of an explanation. It tells you the business has been good. It tells you nothing about whether it will stay good, which is the only thing you are actually being paid to have a view on.

Worse, it fails in exactly the situations where it matters most. A company at the peak of a product cycle looks identical, on returns on capital, to a company with a structural advantage. You find out which one you owned about four years later.

A stricter test

The question I try to answer instead is not "is this hard to attack" but something narrower:

The distinction is the whole thing. Difficult means slow. A determined attacker with real money will accept slow. Irrational means the attacker runs the numbers, sees that the attack destroys value even if it succeeds, and walks away. That is a moat. The first is a head start.

The practical version of the test: what would it cost, in money and in time, to reproduce this position from scratch, and does the reward justify it? Money compresses most things. It does not compress everything, and the exceptions are where durable advantage lives.

What passes

Physical scarcity. Some things cannot be manufactured at any price. Land in a specific place with a specific interconnection. Water rights. A permitted site. An attacker with unlimited capital cannot conjure more of it, and the incumbent does not have to be clever to keep it. This is the least glamorous category and the most durable.

Capital plus time plus accumulated know-how, together. Any one of the three is buyable. All three at once is not. A leading-edge semiconductor fab needs tens of billions of dollars, several years, and a yield learning curve that only exists inside people who have already done it. An attacker can fund it and still be four years behind, by which point the target has moved again. The time component is what makes this category work, and it is why capital intensity alone is not a moat: plenty of capital-intensive businesses earn nothing.

Shared scale economies. The structure where the incumbent passes its scale advantage to customers as lower prices rather than keeping it as margin. This is unusually hostile to attackers, because the attacker must match the price without the scale, which means losing money on every sale, indefinitely, with no visible path to parity. The incumbent's low margin is not a weakness. It is the defense.

Counter-positioning. The incumbent could copy you and will not, because doing so destroys something it already owns that is worth more. This is the strongest short-term protection and the least permanent, because eventually the thing being protected declines in value enough that copying becomes rational. It buys years, not decades, and it should be underwritten that way.

Switching costs measured in other people's time. Not contractual lock-in, which expires, but the cost of retraining hundreds of people and rebuilding workflows that currently work. The person who would have to approve that migration is not the person who would have to do it, which is precisely why it does not happen. Note the honest limit: this protects the installed base and does very little for the next sale.

What fails, despite looking like it passes

Being first. First is a fact about the past. It confers advantage only where it converted into one of the categories above. Plenty of first movers were simply the first to demonstrate that a market existed, at their own expense, for someone else's benefit.

A technology lead. The most commonly mistaken one. A lead is a stock of advantage that depletes unless continuously refilled, and the refill costs money every year forever. Where the lead is genuinely durable it is usually because of the accumulated know-how underneath it, not the technology itself, and it is worth being precise about which one you are actually buying.

Brand, on its own. Brand is real, and it is a maintenance obligation rather than an asset. Stop spending and it decays on a schedule. Brand combined with distribution or scale can be formidable. Brand alone is a rented advantage with the rent due annually.

A long customer list. Revenue concentration among large customers gets presented as a moat and is frequently the opposite: a small number of counterparties who know exactly how much you need them.

Low prices without a structural cost advantage. If you are cheapest because you have chosen to be, a competitor can also choose. If you are cheapest because your cost base is genuinely lower, that is a different sentence with a different meaning.

How moats actually die

Here is the part that makes durability hard to underwrite. Moats are rarely stormed. The ground moves.

Almost nobody loses to a competitor doing the same thing slightly better. They lose because the thing being defended stopped being the thing that mattered. The moat was around a castle, and the trade route relocated. Distribution advantages evaporated when distribution stopped being scarce. Scale in manufacturing mattered less when the binding constraint moved to design. The incumbent's advantage was not defeated; it was bypassed.

This has a specific implication for how to hold a position. The question to revisit is not "is the moat holding," which will read as yes right up until it does not. It is "is the thing this moat protects still the scarce input?" That question has a different answer at different times, and it usually changes quietly.

It also explains why the physical-scarcity category ages best. A moat around a scarce physical input survives the trade route moving, because the new route still needs power, land, and water.

What this does to the coverage board

This test explains most of what gets written here, and it is worth making the connection explicit rather than leaving readers to infer it.

The businesses I have covered are, with one exception, positioned on things capital cannot compress quickly: a manufacturing position that takes years and tens of billions to approach, acreage that cannot be manufactured at all, a membership model whose thin margin is the defense, a two-sided marketplace whose supply relationships took decades to assemble. That is not a coincidence in selection. It is the filter.

The exception is Palantir, which sits in the switching-cost category, the one I trust least over long horizons and said so in the report. Its economics are currently extraordinary. Whether the advantage is structural or is a very good product in a fast-moving market is, honestly, not something I can settle from the outside yet.

And a moat passing this test tells you nothing about what to pay for it, which is the subject of a separate note and the reason several of these companies carry a HOLD.

What would prove me wrong

• The physical-scarcity category underperforms the others. If, over a full cycle, businesses built on irreplaceable physical assets deliver worse returns than those built on brand or technology leads, then I have confused durability with dullness and I am paying for permanence that the market already prices correctly.
• Counter-positioning proves more durable than I assume. I treat it as buying years. If incumbents reliably fail to respond even after the economics clearly favor responding, the category deserves more weight than I give it.
• The irrationality test proves unfalsifiable in practice. The honest worry about my own framework is that "would a rational attacker walk away" is only answerable after the fact, which would make it a story rather than a test. If I cannot state in advance what would count as a failed attack, the framework is doing less work than I think.

That last one is the one I would attack first if I were reading this rather than writing it.