Alibaba's USD 1.5 billion offer for Pupu is usually read as a land grab: more dark stores, more neighbourhoods covered. As of November 9, 2026 regulators were still reviewing the deal, so the number that will decide whether the price was right is not the store count — it is how often an item is missing when a rider arrives; and if a stockout prices a catchment more honestly than rent does, what exactly is the bid buying?
What the bid actually buys
A dark store is worth what its catchment orders, and a catchment orders what it can actually get. That is why availability, not floor space, is the honest unit of account. Two networks with the same number of sites can differ by a factor of two in order density simply because one keeps its top items in stock and the other does not, and the difference shows up in repeat purchase before it shows up in any acquisition multiple.
The bid, in other words, buys a set of shelves inside a set of rings, and the shelves are only worth what they can keep in stock. Whether those shelves convert into a business depends on replenishment discipline that no purchase agreement transfers automatically, which is why the first quarter after closing usually looks better than the second.
Stockout is the unit that prices a network
A bought store still has to be stocked
Acquiring sites shortens the time needed to reach a ring, and does nothing for the time needed to restock a shelf. In an instant retail model, the pick face is the store itself, so an item missing at 8pm cannot be recovered from a regional warehouse: there is no time. Availability therefore has to be managed store by store and hour by hour, which is an operating habit rather than an asset.
This is where acquisitions disappoint. The buyer inherits leases, staff and locations; it does not inherit the daily discipline of counting the top two hundred items before the evening peak. Networks that treat the deal as capacity tend to see order density rise for a quarter and then flatten, because the binding constraint moved back to the shelf.
Catchment beats square meters
A three-kilometre ring contains a fixed number of households, and those households order a certain number of times per week. Divide the fixed cost of the site by that order count and the cost per order falls out. Add square meters without adding households and the same fixed cost is spread over the same orders, which is why an acquisition that looks cheap per square metre can be expensive per order.
The same arithmetic explains why store formats differ by city rather than by company strategy. A dense downtown ring supports a smaller site with a narrow assortment; a suburban ring needs a wider assortment to reach the same basket size, and a wider assortment is exactly what raises the stockout risk that the model cannot absorb. Sortiment breadth, in this model, is a cost that shows up as a missing item rather than as inventory.
Three ways to price a network
| Method | What it measures | Where it misleads | Signal to check |
|---|---|---|---|
| Cost per square metre | Site economics | Ignores household count in the ring | Orders per site per day |
| Cost per order | Density and fixed cost together | Hides the delivery leg | Cost per delivered order |
| Cost per repeat customer | Availability and experience | Slow to move, lags by a quarter | Repeat purchase share |
The third method is the slowest and the most honest. A network that keeps its best-selling items available earns a second order; one that does not, does not. Since the acquisition price is paid once and the repeat order compounds, the availability number is the one that decides whether the multiple was reasonable.
What to check before the deal closes
The four checks below can be run on public disclosures and store visits, without access to the seller's systems, and each one tests whether the ring is real or only drawn on a map. They are ordered by how quickly a wrong answer shows up, because a network that fails the first check rarely passes the fourth. None of them requires the seller's cooperation, which matters while the regulatory review is still open.
- Count households inside each ring, then compare with the number of orders the site handles per day, rather than with revenue.
- Visit the top twenty items in each acquired site at the evening peak and record availability, because that is the hour the model is paid for.
- Read the lease terms for the shortest remaining contract in each cluster, since a site that must be renegotiated next year is priced differently.
- Ask for the repeat purchase share by cohort, and treat any answer given as a city average as incomplete.
| Field | Example value | Reading |
|---|---|---|
| Households in ring | 42,000 within 3 km | Sets the ceiling on order density |
| Orders per site per day | 1,150 | Below 900 the fixed cost dominates |
| Evening availability | 88% of the top 200 items | Under 90% repeat orders stall |
| Repeat purchase share | 34% in month two | The cleanest sign the ring works |
Run the four together and the acquisition stops being a story about coverage. It becomes a question about whether the shelves inside the rings can hold the assortment the catchment expects, which is answerable before the closing and not after.
FAQ
Does buying a dark-store network buy density?
It buys sites inside rings. Density still depends on how many households order and how often, which is why availability matters more than store count.
Why is evening availability the number to test?
Because that is when demand peaks and when a missing item cannot be recovered from a warehouse. A store that fails at 8pm loses the order entirely.
How does a stockout affect the acquisition multiple?
Through repeat purchase. Missing items cut the second order, and the second order is what pays for the fixed cost of the site.
Should the deal be judged on revenue growth?
No. Revenue growth was already running at 45% before the deal. What the deal changes is whether the acquired rings reach density sooner than building would.
