Editorial Article
8,400 Dark Stores and Not One of Them Is Getting Busier
The world’s operator-controlled dark-store estate processed USD 28.55 billion of order value in 2025 across roughly 8,400 sites — USD 9,312 a site a day — and it will reach USD 63.51 billion by 2033. Along the way, the average store gets quieter.
Orders rise from 2,325 million to 3,849 million across the forecast, 6.5% a year. Site count rises from 8,400 to 14,500, 7.06% a year. Orders per site per day therefore fall from 758 to 727. That is not a rounding effect or a transitional year; it holds across the whole period, because new sites are opened in thinner catchments than the ones already trading.
The decomposition is exact. The 10.51% value growth is 7.06% more boxes multiplied by 3.22% more value per box. That 3.22% is itself a 3.76% rise in the average basket offset by a 0.52% annual decline in orders per site. Two forces, working in opposite directions, and the net is a market that grows because it builds.
What that means if you sell into it
A market whose value growth comes from node count has a very specific demand signature. Refrigeration, shelving, shop-fitting, pick-cart and site-security demand tracks the 7.06% node rate, not the 10.51% value rate. Software and automation demand tracks throughput per site — which is falling — and comes in at 3.22%. Those are materially different growth rates for two halves of the same supplier base, and getting them the wrong way round is an easy way to mis-size a market.
It also means the growth lever is site selection rather than store productivity. An operator who wants more revenue from the estate as it stands has to raise basket, because the other term is going the wrong way. An operator who wants more revenue full stop opens more boxes and accepts that each one is, on average, less busy than the fleet.
The average is hiding everything
758 orders per site per day is a blend across formats that differ by more than an order of magnitude, and it is worth being blunt about how little the average means.
| Operator | Format | Orders per site per day |
|---|---|---|
| Rohlik | Large-format automated fulfilment centre (9 sites) | ~6,910 |
| Zepto | Indian dark store | 2,140 |
| Blinkit | Indian dark store | 1,357 |
| Swiggy Instamart | Indian dark store | 1,093 |
| Instacart | No nodes at all; picks from other retailers’ shelves | 0 |
India holds 4,525 of the 8,400 nodes — 53.9% of the global estate by count — which is why the Indian operators set the character of the average. China’s franchise shop-warehouse layer is excluded from the count entirely, because those are independently owned businesses rather than operator-controlled sites. And Instacart is in this market with 338.8 million orders and USD 37.2 billion of gross transaction value while running no boxes whatsoever.
The property business, stated plainly
None of this is a criticism of the format. It is a description of what the format is. A dark store is a small, cheap, fast-to-open piece of urban logistics real estate whose economics are set at the moment the catchment is chosen. The sector’s aggregate behaviour — more sites, each slightly less busy — is exactly what that description predicts when operators are competing for the best catchments and then moving outward.
The consequence is that the number to watch is not the average across the estate. It is the spread, and specifically whether the tail of thin-catchment sites being opened in 2031 looks like the sites opened in 2025. If it does not, the 727 is optimistic. If it does, the forecast holds — and the growth really is a property business measured in square metres and refrigeration.
The estate is counted on a controlled-fulfilment basis: dark stores, micro-fulfilment centres and operator-held shop-warehouses. Operator throughput figures are for the quarter to 31 March 2026, as disclosed. Full definitions and per-format economics in the Dark Store Networks & Micro-Fulfilment Market report.
Sources
Every node count, order count and orders-per-node figure in this article is computed from a disclosed store count, a disclosed order value and the actual number of days in the period:
- Eternal Limited, quarterly shareholders’ letters — Blinkit net order value, store counts and orders.
- Swiggy Limited, financial results and shareholders’ letters — Instamart store count, order value and quarterly loss.
- Zepto Limited, UDRHP-I filed with SEBI, June 2026 — store count, net order value and per-order loss.
- Instacart (Maplebear Inc.), Form 10-Q for the quarter ended 30 September 2026 — gross transaction value and order count for the US benchmark.
The global node count of 8,400 and the 3.22% value-per-site residual are Pheonix estimates built from those national disclosures, since no source publishes a global dark-store estate count.
Frequently Asked Questions
Why does the average dark store get less busy over the forecast?
Orders rise 6.5% a year and site count rises 7.06% a year, so orders per site per day fall from 758 to 727 across the forecast. New sites are opened in thinner catchments than the existing average, which pulls the average down even as total order volume grows.
How is the 10.51% growth rate composed?
The estate was USD 28.55 billion in 2025 across about 8,400 sites and reaches USD 63.51 billion across 14,500 sites in 2033, a 10.51% compound rate. That decomposes into 7.06% more sites multiplied by 3.22% more value per site, and the 3.22% is a 3.76% basket rise offset by the 0.52% annual fall in orders per site.
Why is the global average of 758 orders per site per day misleading?
Because the formats differ by more than an order of magnitude. Rohlik's large-format automated centres run about 6,910 orders a site a day, Indian dark stores 1,093 to 2,140, and Instacart runs no nodes at all while moving 338.8 million orders. A single global average blends all of those and should never be used as a benchmark for an individual site.
What does node-count growth mean for equipment and automation suppliers?
Refrigeration, shelving, shop-fitting and site-security demand tracks node count at 7.06%. Software and automation demand tracks throughput per site, which is rising at 3.22%. Those are two halves of the same supplier base growing at materially different rates, and applying the value growth rate of 10.51% to either overstates the market.
