Cellar Partner Memo

The bet

We are not betting on a marketplace. We are betting on a single claim about behavior: that collectors who store with Cellar keep transacting through Cellar even after the novelty wears off and a competitor offers a lower fee.

If they stay because the custody and grading genuinely lower the friction and risk of every future trade, we are early to an infrastructure business with switching costs measured in physical inventory. If they stay only until someone undercuts the take rate, we are funding a warehouse with a website, and the margin compresses the moment the category gets competitive. Repeat-buyer rates, GMV, attach rate, all of it is downstream of that one question, and none of it can be read until you've answered it, because the early numbers look identical under both stories.

This is the trap the deal sets: it looks like a bet on a charming category, and a charming category attracts enough early enthusiasts to manufacture good cohort numbers whether or not the switching cost is real.

Why consensus is worthless here

The bull and the bear agree on every observable: GMV is growing, repeat purchase is healthy, custody is attaching. That agreement is the tell that the observables don't decide this. Each of those is a level a young, enthusiast-heavy marketplace produces for free, and each is equally consistent with a durable trust platform and a temporary hobbyist clubhouse. The only quantity that separates them is a derivative: how retention behaves when a cheaper alternative appears and the early-adopter glow fades. Underwriting the levels is how we end up paying for enthusiasm and calling it infrastructure.

The mispricing

Two camps are circling and they are wrong in opposite directions. The consumer-marketplace funds price the listing flywheel and underweight whether custody is a real margin engine or a subsidized loss-leader. The alt-asset and fintech funds price the "wine as an asset class" story and underweight whether a genuine two-sided consumer market exists underneath it. Each is discounting Cellar for the exact thing the other should be paying up for.

The company neither camp is underwriting is the real one: a custody-and-authentication utility for a high-value collectible category, where the physical inventory under management is the switching cost and the marketplace is the demand layer that feeds it. Because nobody prices that entity, it likely clears at the lower of two wrong numbers rather than a blended right one. Whoever underwrites custody, marketplace, and authentication trust as one system is buying a more defensible business than either camp thinks it's pricing. This is the entry point, and it has a half-life: the day a sharp consumer fund and a sharp alt-asset fund compare notes, the gap closes.

The one diligence question

That comparison, and only that comparison, adjudicates the bet. If stored collectors keep transacting at materially higher rates and don't flee on price, custody is a real switching cost and we are looking at infrastructure. If stored and unstored behave the same once the novelty fades, the warehouse is a cost we're capitalizing for no durable return. No other metric separates the trust platform from the hobbyist clubhouse, and the two are not close in value. There's a tell inside the answer, too: if they aren't cohorting stored versus unstored behavior at all, they are managing to GMV rather than to the thing that determines whether GMV persists.

What turns this into a lead check

A custody cohort that retains and re-lists at a durable premium to the transaction-only cohort, and holds that premium where a cheaper competitor exists. That is proof the moat is the inventory, not the novelty, and it converts me on its own. Custody running at or above breakeven on its own economics, and repeat-buyer breadth widening rather than concentrating, would harden it, but they only matter if the custody premium holds first.

Why this could be wrong

The strongest bear case isn't that the company is bad. It's that we'd be telling ourselves the infrastructure story because the cohort chart is flattering, when the same chart fits the hobbyist-clubhouse version exactly.

The demand may be a thin layer of wealthy enthusiasts. High-value collectible markets concentrate; a marketplace can look healthy on the backs of a few dozen serious collectors and have no path to the breadth a venture outcome needs. If repeat purchase is deepening on whales rather than widening across buyers, the ceiling is far lower than the GMV curve suggests, and that failure mode reads as success right up until growth stalls.

The custody moat may be shallower than it looks. Storage and authentication are services an incumbent auction house or a well-capitalized retailer can stand up if the prize is large enough; the switching cost holds only while Cellar is too small to be worth attacking. Make the category attractive and someone with existing logistics and a trusted brand builds the warehouse, and the moat fails at the moment it's first tested.

The category may simply be too small or too slow. Fine wine is large in dollars but thin in transaction frequency; a collector trades rarely, and a marketplace monetizing infrequent high-value trades can struggle to compound the way a high-frequency marketplace does. From the outside today, "infrastructure with switching costs" and "a beautiful low-frequency niche" look identical. The custody cohort comparison is what tells them apart.

View

Lean in, conditional on one comparison, and treat the condition as the whole decision. Not "interesting, stay close": this is a lead check or a pass, and the stored-versus- unstored cohort behavior tells us which. If the custody cohort retains at a durable premium and holds it against cheaper competition, we are looking at a defensible business that two camps are each mispricing, and we should lead before they compare notes and the gap closes. If the cohorts converge once the novelty fades, it is a lovely company in a slow category, and we pass clean. Everything rides on that premium, so we do not write a check until we have seen it.