What a dark kitchen costs: the before and after numbers

The real CAPEX: 12,000 to 45,000 USD depending on where you build it

A dark kitchen costs between 12,000 and 45,000 USD in initial investment, and what moves that figure three and a half times is not equipment quality but the choice between entering a shared hub or building out a raw shell. In a hub cubicle with hood, grease trap and electrical points already installed, you get in with 12,000 to 18,000 USD: cooking line, refrigeration, steel tables, POS and working capital for the first eight weeks. In a raw shell the cost jumps, because extraction, gas and health-code adaptation add 15,000 to 22,000 USD before a single pan arrives. With roughly 7,606 ghost kitchens operating in the United States (OysterLink, 2025), the market no longer rewards whoever opens cheapest; it rewards whoever calculates correctly how long it takes to recover what was put in.

Platform commission: the 42% that leaves before the first kilo of chicken

Monthly operating cost in a dark kitchen eats 28% to 41% of gross sales before touching food cost, and the heaviest block inside that range is marketplace commission. In the P&L I am reviewing as I write this —47 million pesos of sales in July 2026— commission weighed 12.9 million, in-app advertising 4.1 million and the hub cubicle rent 2.8 million: 42% of billing handed over before buying a single input. With food cost sitting in the optimal 28-35% band set by the National Restaurant Association, the arithmetic closes negative without anyone having made a kitchen mistake. What breaks these projects is never a badly costed recipe; it is a cost structure that punishes every additional peso of sales at exactly the same percentage.

Market size explains why commission is not coming down

Platforms have zero incentive to lower your commission, and the 2024 figures say so without ambiguity. Delivery Hero closed the year with group GMV of 48.8 billion euros, up 8% (Delivery Hero, Q4 and FY 2024 results); Just Eat Takeaway moved 8 billion euros of GTV in Northern Europe alone, growing 4% in constant currency (Just Eat Takeaway.com, 2024); Southeast Asian delivery spending reached 19.3 billion dollars after a 13% jump (Momentum Works, 2024). A business of that volume, on that curve, does not negotiate terms with a twenty-square-meter kitchen. Those three figures together trigger one decision: if your financial plan assumes commission improves with volume, the plan is wrong from its very first line.

Marketplace versus owned channel: proportional tax versus fixed cost

Selling twice as much on an app costs you exactly twice as much; selling twice as much through your own site costs practically the same, and that is the structural difference between a dark kitchen that scales and one that merely gets bigger. Commission is a proportional tax that never dilutes, while the owned channel —domain, payment gateway, WhatsApp Business, one contracted rider— is a fixed cost that every additional order makes cheaper. Add the data question: when the order comes through the marketplace, the customer belongs to the marketplace, and you pay again next month to reach them. Deliveroo reported a record frequency of 3.5 monthly orders per consumer across the UK and Ireland during 2024, a figure that in your own channel means one paid order and 2.5 free ones, and on a marketplace means 3.5 commissions. That number decides your model.

Here I was wrong for years: cheap to enter, brutally expensive to hold

For years I recommended the dark kitchen as the cheap way to test a virtual brand, and on CAPEX it is —I still hold that— but it is the most expensive way to buy customers if you do not build an owned channel from week one. That asymmetry between a low entry cost and a very high cost of staying alive is what breaks projects that technically worked. The February 2026 P&L shows it: the day the platform moved his position in the lunch category, sales fell 31% in eleven days, with nothing changing inside the kitchen. At Masterestaurant we cost every dark kitchen under two mandatory scenarios, one with platform sales cut in half, because the algorithm rewards preparation time, acceptance rate and rating, and it raises the advertising bid every time a competitor opens three blocks away.

Last-mile cost is tightening too

If your plan involves leaving the marketplace and running your own delivery, look first at what is happening to rider income, because it sets the floor of what hiring will cost you. Gridwise measured average earnings of 14.96 dollars per hour on Uber Eats in 2024, down 5% year over year, and 12.23 dollars per hour on DoorDash, down 3%. DoorDash itself generated more than 18 billion dollars in earnings for riders during 2024 (DoorDash, full year results). Translated into your operation: rider supply exists and is plentiful, yet hourly income is falling, and an underpaid rider hands you back delays, cancellations and low ratings. The decision: budget your own delivery above the market average, never at the average, or do not build it at all.

Foodtech, capital, and the cash flow that actually kills you

Delivery and dark kitchens sit among the most funded verticals in Latin America (Bloomberg Línea), and that abundance of capital has a perverse effect on the owner opening with their own money: you compete against operators who can sell below cost for eighteen months. Meanwhile, cash flow remains the leading cause of financial stress and closure among small businesses (Inc.), and a dark kitchen concentrates the three classic aggravating factors: platform payouts at fifteen or thirty days, inputs paid cash or at eight days, and an advertising spike paid in advance. A dining-room restaurant collects on the spot; you do not. Before signing the hub contract, calculate how many weeks of payroll and inputs your current cash covers, and if the answer drops below eight, do not open yet.

The 3 figures you should tattoo on the wall

Three numbers and what to do with each. First: 28-41%, what structure eats from your gross sales before food cost —action: cost your menu against the NET price left after commission, never against the app price, and raise platform prices enough to match the direct channel's margin. Second: 3.5 orders per month per active consumer (Deliveroo, UK and Ireland, 2024) —action: set up phone capture and WhatsApp ordering from day one, because recovering that same customer four times a month through your own channel is the only lever that gives margin points back. Third: 28-35% food cost (National Restaurant Association) —action: measure yours weekly per dish, not monthly in aggregate, and pull from the digital menu everything above 35% once the net price is already discounted.

The three differences that move cash

The first is structural rather than a matter of effort: marketplace commission is a proportional tax that never falls with volume, while the cost of an owned channel is fixed and dilutes. Doubling sales on Rappi costs you exactly double; doubling sales through your own site costs roughly the same, and that gap between proportional and fixed is what separates a dark kitchen that scales from one that merely gets bigger. Second, the data. When the order comes through the marketplace, the customer belongs to the marketplace, and you pay again to reach that same person next month. With an owned channel, the second order from the same buyer carries zero acquisition cost, and since average frequency for prepared-food delivery sits near 2.3 orders per month among active users, the compounded saving over a year is the difference between covering rent and not covering it.

The three differences that move cash — in practice

Third, the one almost nobody measures: ticket size. Inside the app your product competes on a grid beside nine rivals with photo and price visible, so the user anchors on the cheapest. On your own menu, without that grid, average ticket climbs from 8.90 to 11.40 USD because suggestive selling works — drink, dessert, double portion — and because nobody is comparing you to the neighbor on the same screen. There is a genuine tension here and I will not hide it: the marketplace gives you volume from day one, and an owned channel takes four to seven months to matter. Anyone who abandons Rappi to "save the commission" goes broke before their site has traffic. The resolution is sequential, not ideological: use the marketplace as paid acquisition, capture the customer inside the packaging, and move repeat purchase to your own channel. The marketplace is your top of funnel; your site is the bottom.