How to start a dark kitchen from scratch: the expensive mistakes vs the right method

What does it actually cost to build a dark kitchen from scratch?

Building a dark kitchen from scratch costs between 18,000 and 65,000 USD as of September 2026, and the range runs that wide because three different businesses fit inside it.

An operator in Medellín sent me his P&L in March: 41 million pesos in sales against 38 million in costs, eleven months of operation, and less than 20% of the initial investment recovered, with sales climbing month after month. They were climbing, true. So was the platform commission, because he had signed at 30% to get into the promo carousel and never sat down to renegotiate. Capital was not the problem in that statement; the problem was arithmetic nobody did before signing. And this sector forgives nothing: ghost kitchens already accounted for roughly 15% of U.S. foodservice delivery sales in 2023, according to Statista, with all the competitive pressure that drags along. Each range buys something different, and you had better know which one before the money moves.

What each investment range buys you?

Between 18,000 and 26,000 USD you get into a shared kitchen:

a monthly fee of 900 to 2,200 USD depending on the city, your own light equipment (double fryer, griddle, two-door refrigeration, hood already installed by the hub operator), 2,500 to 4,000 in smallwares and opening packaging, health permits, and around 3,000 of working capital. The 27,000 to 44,000 bracket means a small dedicated unit, 40 to 70 square meters: shell work and electrical adaptation between 9,000 and 16,000, grease trap, extraction, plus the rent deposit almost nobody budgets. Above 45,000 we are talking about a kitchen with two or three production lines and capacity for three virtual brands at once, with a blast chiller, larger cold storage, and a POS integrated to the aggregators. That last bracket only makes sense if your demand is measured, not projected.

The arithmetic error that kills projects that still have cash

Most projects that die did not run out of capital: they handed the price of their product to a third party that optimizes its own business. Run the math with me on a 12 USD ticket at 30% commission. You keep 8.40 USD. Food cost at 32%, which is the CEILING and not a target, takes 3.84 USD. Decent packaging, with a lid that does not spill and a tamper seal, costs 1.20. That leaves 3.36 USD per order to pay kitchen labor, utilities, rent, and your profit. If your kitchen payroll and fixed costs add up to 6,000 USD a month, you need 1,786 monthly orders just to break even, roughly 60 a day, every single day, without one slow Monday. You have to know that number BEFORE signing the aggregator contract, because afterward it stops being a negotiation and becomes survival. Five variables explain almost the whole gap between 18,000 and 65,000, and I rank them by real impact on your number.

Five factors that move the final price

Extraction and the grease trap weigh 4,000 to 11,000 USD when the space was not already fitted out, and that line item blows up more budgets than any other. Three-phase power, when you have to file for it and run the cable, adds 1,800 to 5,000 depending on the distance to the transformer. Your menu decides the equipment: a card with frying, oven, and griddle triples the investment against a cold-assembly card, and there sits a 6,000 to 14,000 USD difference. The city swings rent by a factor of three between a secondary zone and a restaurant corridor. And working capital, which hardly anyone counts as investment, should cover ninety days of full operation, between 4,000 and 9,000 USD. Let me take an unpopular position: shared kitchens are excellent for validating and terrible for scaling, and mixing that up turns a cheap test into an expensive dependency.

Shared kitchen or your own unit: these are not stages of one thing

Inside a shared hub you pay for usage and your structure is almost entirely variable, which is exactly what you want while you find out whether your menu sells. A dedicated unit is not the next level of the same game: it is another business, with rent, utilities, and payroll running fixed whether or not a single order comes in. What happens if you jump to your own unit on the demand of one good Tuesday? You sign thirty-six months, build capacity for 120 daily orders, run 55, then cut prices to fill the kitchen, damaging the very margin that justified the jump, and you end up tied to a lease you can no longer pay. Jump when sustained volume beats the hub's variable cost, never before. Four negotiating levers with the aggregators actually move the needle, and none of them is politely asking for a discount. First: set a volume floor and ask for a quarterly rate review in writing, because platforms adjust commission once you prove incremental orders.

How to negotiate with the aggregators and cut your real cost?

Second: separate the logistics fee from the marketplace fee and negotiate only the one tied to the service you truly use. Third:

build your own channel in parallel from month one, with WhatsApp and direct ordering, even if it only carries 15% of volume, because that fraction gives you something to show at the table. Fourth: raise the ticket with high-margin add-ons before raising the price of the main dish. Diego F. Parra keeps hammering one order at Masterestaurant that sounds obvious and almost nobody follows: first you learn your contribution margin per order, then you sign. More than 25% of operators already use artificial intelligence, according to the National Restaurant Association, and its best use today is watching that margin daily, not writing menu descriptions. Delivery stopped being an add-on and became the infrastructure of the business, and that changes who you compete against. Some 65% of limited-service operators already offer delivery, according to the National Restaurant Association, so supply inside every app multiplied.

The channel numbers that explain the competition waiting for you

DoorDash closed 2024 with roughly 2,583 million orders and 685 million in the fourth quarter alone, up 19% year over year, per its own results. DiDi Food has delivered more than 360 million orders in Mexico across five years. In China, Meituan and Ele.me clear more than 60 million daily orders according to Mordor Intelligence. Translated to your screen: in several Latin American cities a search for burger returns more than 200 results within a three-kilometer radius. You are not competing against the restaurant on the corner, you compete against infinite scroll, and there the photo, the prep time, and the review win, in that order. Start by tracking three daily numbers and park everything else, because scattered attention kills more dark kitchens than competition does.

The first ninety days: what to measure and in what order

The first is contribution margin per order, net of commission and packaging, calculated per dish rather than averaged; the second is the time from order entry to the moment it leaves the door, because one extra minute in the kitchen turns into one less star; the third is the share of orders coming through your own channel. With those three numbers you decide whether to raise a price, pull a dish, or switch off a promotion. One figure to size up the urgency: close to 70% of operators planned to invest in technology over the following year, according to the National Restaurant Association, and 48% put point of sale first. Close month three having pulled one dish from the menu for negative margin. If you found none, you are not measuring dish by dish. The difference sits not in equipment or premises but in WHICH decision comes first. Sign with aggregators before knowing your contribution margin per order and you hand pricing power to a third party, one that optimizes its own business rather than yours.

Where the two routes really split?

A 30% commission on a USD 12 ticket leaves USD 8.40 to cover food cost, packaging, kitchen payroll, utilities and profit.

At 32% food cost (USD 3.84) plus USD 1.20 packaging, USD 3.36 per order remains for everything else. Now you know how many orders you need. An owned unit is not an advanced stage of the shared kitchen: it is a different business with a different cost structure. Shared space means you pay per use and almost everything stays variable; owned premises turn rent, utilities and base payroll into fixed obligations that demand volume every single month, rain or shine. Moving between them before reaching 900-1,200 stable monthly orders trades flexibility for leverage at the worst possible moment. The virtual brand is the lever almost nobody pulls correctly. One production line can carry two or three brands when they share 70% of ingredients, and each brand surfaces in different searches inside the aggregator.

Where the two routes really split — in practice?

Operators running three brands on a single kitchen report 40-55% more orders at the same fixed cost, because kitchen occupancy during valley hours rises.

The condition without which none of this works: all three brands must share technique and equipment, or you have just built three kitchens in one room. Here is the counterintuitive part: cutting prices inside the aggregator rarely lifts profit. Platforms push promotions because promotions grow their GMV, and the operator who joins a 2-for-1 raises orders while lowering margin at the same time. Better to raise average ticket through built combos —which lift perceived value without touching percentage food cost— than to chase cheap volume in a saturated category. If your dark kitchen also handles pickup or runs a window, the PHYSICAL menu still matters. The QR menu solves price updates, delivery and analytics, while the printed menu controls the pace of the decision and enables suggestive selling at the counter. Both, each in its role; removing the physical one to «modernize» strips away the single tool that steers a guest's eye toward your highest-margin plate.