Shared vs own cloud kitchen: the call that fixes your margin before the first order

1. Which one works if you bill under USD 500K a year in digital

Below USD 500,000 a year in the digital channel, the shared kitchen wins outright, because it lets you cancel the commitment in 60 days while your own kitchen ties you to a five-year lease and to an investment you never recover if the concept misses. At that volume you move between 1,800 and 3,000 orders a month; with the USD 18 average ticket I see on most delivery dashboards across the region, that means USD 32,000 to 54,000 in monthly sales, and no industry data set justifies sinking capex against that base. Margin rules here: high-performing ghost kitchens run between 10% and 30% according to OysterLink (2025), against the 3% to 5% of the traditional restaurant, and that gap gets built with variable cost, not with square meters you own. Between USD 500,000 and 1 million a year the call no longer resolves with a rule, it resolves by measuring module occupancy.

2. The USD 500K to 1M band: where it stops being obvious

If your brand uses the shared module less than 65% of usable kitchen time, keep renting; above that threshold the peak-hour surcharge starts eating the savings and it makes sense to open the own-kitchen conversation with a partner who puts up the bricks. In this band the typical operator runs 2,800 to 5,500 monthly orders and holds a contribution margin of 22% to 28% per order, as long as the aggregator commission stays below 30%. The market gives you time to think: cloud kitchens are projected at USD 83.5 billion in 2026 with a 9.7% compound rate through 2034 according to Fortune Business Insights (2026). Once the digital channel clears USD 1 million a year and demand concentrates into windows —Friday and Saturday from seven to ten at night generating more than 40% of weekly sales—, your own kitchen stops being a luxury and becomes defense.

3. Past the million mark, with concentrated demand, the owned kitchen pays for itself

That profile implies roughly 4,600 monthly orders, and a shared module with six brands fighting over the same exhaust hood produces dispatch times the aggregator algorithm punishes with less visibility. Less visibility next week means fewer orders, and the loop feeds itself until the brand vanishes from the first screen. Diego F. Parra puts it this way inside the Masterestaurant framework: you are not buying square meters, you are buying peak capacity. The scale figure backs the concentrated-volume logic: DoorDash's marketplace moved USD 21.3 billion in the fourth quarter of 2024, up 21% (DoorDash, 2024). Above USD 5 million a year a different profile shows up —the media-chef brand, the gastronomic celebrity, the large-format themed concept— and there the right answer is almost always hybrid: an owned kitchen as production core plus rented shared modules in the cities where demand is still unproven. United States data supports that design: 86.9% of virtual brands operate a hybrid model and barely 13.1% live exclusively online, according to Locmatic (2024).

4. Above USD 5M: the celebrity-chef profile and the large-format themed concept

A brand carrying that much reputational weight cannot afford a 48-minute dispatch time on a Saturday, because the cost of the bad review gets paid in the dining-room channel and in licensing deals, not just inside the app. Numeric rule: build the core yourself when a single city clears USD 1.5 million a year. From USD 10 million a year up, a group or chain stops deciding kitchen by kitchen and starts deciding network topology. The architecture I recommend has three layers: owned kitchens in the five to eight cities that concentrate 60% or more of sales, shared modules in test cities, and a satellite finishing kitchen for seasonal peaks. Brooklyn Calzones reached 1,474 locations with 12% share among United States virtual brands according to Locmatic (2024), and it did not get there by laying bricks at every point. Virtuality as an expansion lever already carries weight: 32% of restaurant growth strategies in 2025 run through virtual brands according to Technomic via Apicbase (2025).

5. Groups and chains above USD 10M: the mixed network

A group replicating owned kitchens everywhere is buying territory risk it could have rented. Renting turns fixed cost into variable cost, and that rewrites the financial discipline of the business entirely: you stop needing a minimum volume to survive and start needing a minimum CONTRIBUTION MARGIN per order. Those are two different trades, and plenty of operators run the second with the habits of the first, watching total sales when they should watch the USD 4.80 left over from every USD 18 ticket. With 30% food cost, a 28% aggregator commission and 6% packaging, USD 6.48 per order remain to cover direct labor and the module rent; if the module costs USD 3,500 a month, you need 1,100 orders just to cover it. The whole sector works on 3% to 9% net margins according to Statista, so a half-point mistake does not forgive. Buying your own kitchen means buying a five-year hypothesis about a zone; renting a module means leasing that same hypothesis with the right to hand it back.

6. Territory risk gets bought or rented, and almost nobody models it

The right question is not what the rent costs, but what happens if the zone cools down: if the neighborhood loses 25% of demand because a rival logistics hub opens, the shared-module operator relocates within a quarter while the operator with owned construction carries a plant running at 46% utilization for the whole contract term. That 46% real occupancy is the figure I have most often had to put in front of an owner convinced his problem was paying someone else's rent. Emerging markets amplify the argument: Mexico moved USD 1.1 billion in cloud kitchens during 2024 with 10.74% compound growth through 2033 according to IMARC Group (2024), and at that speed the hot zones keep moving.

7. The decision, one line per band

Rent while the digital channel stays under USD 500,000 a year; measure occupancy between USD 500,000 and 1 million and only move if you pass 65% of usable time; build your own kitchen above the million mark when more than 40% of sales concentrate into two windows; combine both above USD 5 million with an owned core in any city clearing USD 1.5 million; and above USD 10 million design a three-layer network before signing a single construction contract. The sector is not shrinking —the delivery app market closed 2024 at USD 110 billion, up 15.5% according to Business of Apps (2025)—, but that does not oblige you to buy the bricks today. Open your dashboard, calculate your module's real occupancy this week and decide against that number. Shared turns fixed cost into variable, and that rewrites break-even: you stop needing a minimum volume to survive and start needing a minimum contribution margin per order.

8. The four differences an investment committee actually debates

Two different financial disciplines, and plenty of operators run the second one with habits built for the first. Own kitchens hand back control of peak capacity. A shared module with six brands fighting over one extraction hood at nine on a Saturday night produces dispatch times the aggregator algorithm punishes with less visibility, and less visibility means fewer orders the following week. That loop feeds itself, and almost nobody models it. Territory risk is either bought or rented. Shared lets you lease a zone hypothesis for 90 days; your own kitchen buys it for five years. According to Statista Market Insights (2024), online delivery in Colombia moved USD 1.18 billion with a projected 7.32% compound rate through 2029, healthy growth but not explosive, so a badly chosen zone will not be rescued by the market tide. Customer data belongs to whoever controls the last leg. Under a shared roof you rarely see the phone number, the repeat rate or the churn reason; the hub and the aggregator do.

9. The four differences an investment committee actually debates — in practice

That information asymmetry separates a replicable virtual brand from a hot-food supplier with no negotiating power.