How to build a dark kitchen from scratch: the traditional route against the Masterestaurant method

How big is the market you actually want to enter?

Cloud kitchens moved USD 80.3 billion at the close of 2025, and Grand View Research projects USD 88.7 billion for 2026, growing at a 12.6% compound annual rate through 2033.

That huge number fools nearly everyone who reads it, because the global market will not buy a single lunch from you: what decides your cash register is the slice that fits inside a twenty-minute motorcycle radius around the door you are about to lease. Latin America, per the same firm, closed 2024 at USD 12,917.3 million in online delivery with an 8.6% CAGR projected to 2030, and Spain sits near USD 5 billion across delivery and dark kitchens according to Ken Research. Translate before you sign: if your category accounts for 4% of orders in your zone, that global 12.6% growth will not rescue a badly placed warehouse.

The order of decisions is worth more than the price per square meter

You buy DEMAND first and the oven second; reversing that order is the most expensive mistake in this business. The traditional route starts with the lease —space, kitchen, menu, platforms, marketing— and discovers in month four that nobody in that industrial district orders your category, by which point thirty-six months are already signed. The method I use at Masterestaurant opens with the map: what gets ordered within a twenty-minute radius, at what ticket, at what hour; then a short menu, then the platforms, and only then the kitchen. Rearranging that sequence costs nothing. There is accounting behind it too: the lease is the only decision on the list you cannot correct next week, while a bad menu gets rewritten Tuesday and expensive packaging changes with one call to your supplier.

Platform commission is the line almost nobody models before signing

A dark kitchen is not a restaurant without a dining room, it is a logistics business that happens to cook, and the income statement shows it. With a dining room, occupancy costs run high as a share of sales and the guest pays for their own transport; in a hidden kitchen rent drops, but a new line appears that varies by the plan you contracted. You are left with USD 4.80 of contribution margin before payroll, and with that USD 4.80 times the day's orders you pay cooks, utilities, rent and your own profit. Drop the ticket to USD 10 and the same math leaves USD 3.20.

The profitable radius is measured in motorcycle minutes, not kilometers

Seventy cheap square meters on an industrial corridor, with good ceiling height and three-phase power, look perfect on the contract and are a disaster on the map. Physical distance does not rule here: delivery time rules, and past twenty-five minutes the cancellation rate and the one-star review climb hard, which then punishes your position in the listing and dries up your orders exactly when you need them most. Take a plan, mark your door, draw the polygon a courier covers in twenty minutes at peak hour —not at three in the afternoon, which is when everyone runs the test— and count the homes and offices inside it. If that polygon holds less density than the corridor next door where the square meter costs more, the expensive meter turns out cheap.

The algorithm rewards operations, and there the small operator has the edge

Platforms stopped handing out visibility by seniority and now hand it out by operational performance: acceptance time, preparation time, cancellation rate, diner rating. For the traditional method Rappi is a sales channel; I treat it as an algorithm with published rules that rewards consistency, and that difference in reading changes what you look at every morning. A well-run new kitchen overtakes a badly run old brand in the listing within weeks, something unthinkable five years ago. Never before did the small operator hold this much leverage. Almost nobody uses it, because they keep staring at the menu instead of the metrics panel, and because accepting an order in eleven seconds rather than fifty looks like a trivial detail until you see the cumulative effect across thirty days of ranking.

How to read these numbers in YOUR operation: three scenarios?

Benchmarks do not apply the same way across the three sizes, and confusing them wrecks the budget. Small kitchen, a single operator with forty meters and two people:

live off margin per order, demand a USD 12 minimum ticket, and do not launch a second virtual brand until the first one passes sixty daily orders, because splitting attention with that payroll destroys preparation time. Mid-sized operation, eighty to a hundred twenty meters with full shifts: the unit of measure shifts to contribution margin per square meter, and there it does make sense to run two or three brands sharing mise en place. One dataset, three different decisions.

Where these benchmarks come from and how far they reach?

The market-size figures cited here come from public research houses:

Grand View Research for cloud kitchens and for Latin American delivery, Global Growth Insights for dark kitchens (USD 58.1 billion in 2024), Ken Research for Spain, and Statista Market Insights for global delivery, which closed 2024 at USD 1.22 trillion counting groceries and meals. Their limits deserve an honest statement: these are estimates built on differing methodologies, not audits, which is why Verified Market Reports puts virtual restaurants at USD 66.3 billion in 2024 while another house gives a larger number for a similar perimeter. The operating ranges for commission, occupancy and food cost are orders of magnitude from the trade, not sample averages, and you should replace them with your own invoices the moment you have three months of history.

The printed menu stays if there is a pickup window

If your dark kitchen has even a pickup window or a waiting bench, paper stays. People misread this constantly, and the misreading costs them money: the QR code is a complement —it updates prices in seconds, feeds delivery, hands you analytics on what gets viewed and what does not— but the printed menu controls the pace of the order and holds up the suggested sale in front of someone already standing at your counter. Each has its role, and neither replaces the other. Start with what decides the cash register: build the demand map of your twenty-minute radius before visiting a single warehouse, and compare contribution margin per order across the two or three most-ordered categories against the rent you are being quoted. That exercise fits in one afternoon.

Where the two roads genuinely part?

The order of decisions. Traditional: venue, kitchen, menu, platforms, marketing. Masterestaurant: geolocated demand, short menu, platforms, kitchen, marketing. Reordering costs nothing and decides the outcome, because the lease is the one decision you cannot correct the following week.

The unit of measure. The traditional route tracks sales; we track contribution margin per order and per square metre. The relationship with the platform. Traditionally Rappi is a sales channel; for us it is an algorithm with published rules that rewards operating consistency, which is why the real lever to increase sales on Rappi sits in cutting prep time from 22 to 14 minutes rather than in stacking a 25% discount.

Where the two roads genuinely part — in practice?

How the customer is treated. Ghost kitchen owners usually accept that they will never know anyone. We rebuild identity with a packaging insert, a WhatsApp repeat code and a live Maps listing:

in well-run operations that shifts a meaningful share of volume onto the owned channel within a year. The horizon of the menu. A 30-item menu in a dark kitchen is a restaurant menu jammed where it does not fit. Fewer references, faster turns, more virtual brands on one line: that is the gap between USD 4,000 and USD 9,000 of annual sales per square metre.