Restaurante físico o dark kitchen: cuál conviene, priced against your real budget

Physical restaurant or dark kitchen: which one wins?

With less than 40,000 USD in total capital the dark kitchen wins, and above 90,000 USD with a high-footfall street corner the dining room wins;

the middle band is no man's land, and that is exactly where people go under. The reason is not the entry ticket but the STRUCTURE: a hidden kitchen moves its rent burden, which in a street location runs between 12% and 18% of sales, down to a band of 4% to 7%, and hands its margin to the apps instead. An operator in Medellín sent me two quotes on the same Tuesday: 118,000 USD for 90 m² with 42 seats in Provenza, and 26,500 USD for a station in a shared kitchen in El Poblado wired to Rappi, Uber Eats and DiDi Food. Four times cheaper is not four times the business. As of September 2026, across urban Latin America, the investment sorts into three tiers that include very different things.

What each investment range actually buys?

The first, 18,000 to 40,000 USD, buys a station in a shared kitchen: hood and hot line already installed, 6 to 12 usable m², onboarding onto three aggregators, packaging, branding and a three-month operating cushion;

civil works simply do not appear, because the hub operator paid for them. The second tier, 45,000 to 85,000 USD, is a dark kitchen of your own inside a warehouse or second-line unit, and here you do pay for electrical and gas fit-out, extraction, grease traps, health permits and new equipment. The third starts at 90,000 USD with no clear ceiling; that 118,000 USD Provenza quote splits roughly 40% into works and furniture, 25% into the kitchen, 15% into permits and deposits, and the remainder into working capital. The most expensive mistake in this decision is comparing the day-one outlay instead of the cost per order served across 36 months.

Cost per order served beats day-one outlay

A dining room charges its toll up front, through works, furniture and permits; the hidden kitchen charges it every single day for years, through aggregator commissions that moved between 18% and 30% across the region during 2025 and 2026 depending on country and contracted plan. Run the ugly number: 40 daily orders at 12 USD add up to 175,200 USD in annual sales, and a 24% commission takes 42,048 USD every twelve months. Within two and a half years that 26,500 USD kitchen has paid four times its own investment in toll, while the Provenza unit has finished amortizing its build. Anyone who skips that projection decides with half the information. Five variables explain almost all the spread I see between quotes, and each deserves a number. Location rules: moving from a second-line street to a high-traffic corner in Provenza, Palermo or Roma Norte multiplies rent by 2.5 and adds six to ten points of deposits and guarantees.

Five factors that move the price

The condition of the property decides the second block, since a unit that already traded as a restaurant saves between 15,000 and 30,000 USD in extraction, gas and grease traps. Third comes equipment, where buying refurbished cuts the kitchen chapter by 35% to 45% and punishes your warranty. Fourth is seating capacity, because every added chair drags along restrooms, fire exits and air handling. Fifth is input cost, up 35% in the United States since 2019 according to the National Restaurant Association, and enough to push Colombian menu prices 9.8% higher since February 2025 according to ACODRES. The physical restaurant breaks on rent and on the day of the week. A unit signed at 18% of projected sales needs Tuesday and Wednesday full just to breathe, and if those two days barely reach 40% of Friday's sales, the whole operation lives off two shifts and any rainy evening turns into an accounting loss.

Where each model breaks?

The fix is not cutting prices but changing how the room is USED: a set lunch at a 14 to 17 USD ticket, à la carte dinner at 26 to 34 USD, same kitchen and same crew.

The dark kitchen breaks somewhere else entirely, on commission and dependency: when 100% of your orders arrive through three apps, you do not own customers, you rent traffic at 18-30% per transaction, and two extra commission points from Rappi or iFood swallow roughly 15% of your operating profit without asking permission. Say your hidden kitchen bills 180,000 USD a year, keeps 11% operating profit, and the main aggregator —the one bringing 62% of your orders— raises its commission three points and drops your search position because a chain started bidding on that category. First hit: 3.3 margin points, close to 6,000 USD, gone without you touching a single recipe. Second hit: the visibility drop cuts volume by 20%, and since your rent is fixed even though it is low, the break-even point shifts and you slide into losses by month four.

What if the aggregator rewrites the rules next year?

That is the moment people discover they never had a business, they had a ranking position on someone else's platform. The defense gets built BEFORE:

your own ordering channel on your own domain, a customer database and an email list, which opens at around 25.1% on average according to Omnisend's 2024 report and lifts another 26% when the message is personalized according to Stripo. Negotiate the three line items where margin genuinely exists and drop the rest. Rent comes first: ask for three grace months covering the build plus a ladder at 60% of the fee through the first semester, in exchange for a longer term; on a 3,500 USD monthly unit that frees between 10,000 and 14,000 USD of cash in year one, which is precisely the working capital almost everybody lacks. Commission comes second, because aggregators run volume tiers and category exclusivity deals, and whoever clears 900 monthly orders can negotiate two to four points; get it in writing with a review date.

How to negotiate the real bill down?

Then payroll, which together with food makes up half your cost:

AI-assisted scheduling reduces labor cost by 8% to 12% with forecast accuracy above 90% according to TimeForge 2025, and every departure you prevent saves 150% of a salary in replacement cost according to StaffedUp. My criterion, after twenty years inside kitchens and boardrooms across 43 countries, is that the right question is not which model wins but which one survives the mistake you are certainly going to make. Diego F. Parra and the Masterestaurant team settle it with one dry rule: if your total capital does not cover twelve months of operation at 55% of projected capacity, do not open a dining room, open a hidden kitchen and learn to cost with somebody else's money. And if you do open a room, sign the lease below 12% of conservative sales, never optimistic ones, because that percentage point is what separates a business from a badly paid job.

The decision, signed

Take your two quotes today, divide the full 36-month cost by the orders you project serving in that window, and keep the smaller number. That number decides, not your enthusiasm. A dine-in restaurant breaks on rent and on the weekday. A room committed to 18% of projected sales in Provenza, Palermo or Roma Norte has to fill Tuesday and Wednesday to breathe; when those days deliver only 40% of Friday's volume, the whole operation lives off two shifts and any rainstorm turns into an accounting loss. The fix is not discounting, it is repurposing the room: a 14-17 USD executive lunch, a 26-34 USD dinner card, the same kitchen. A dark kitchen breaks on commission and on dependency. When 100% of your orders arrive through three apps, you do not own guests, you rent traffic at 18-30% per transaction. Two extra commission points from Rappi or iFood eat roughly 15% of operating profit overnight, and your only move is raising in-app prices and losing rank to the comparison shopper.

Where each model breaks, and what the break costs?

Residual asset is the least discussed difference and the most expensive one at five years. Close a physical restaurant and you still hold a Maps listing with 400 reviews, a WhatsApp database and a sellable lease.

Close a ghost kitchen and you hold recipes: aggregator ranking does not transfer, does not sell and does not travel to another platform with you. The ramp curve runs opposite to intuition. A virtual restaurant bills from week two because the aggregator already owns the traffic; the physical room needs eight to fourteen months before local SEO and 5★ reviews push on their own. In exchange, dark kitchen sales flatten where the delivery radius ends, typically 4 to 6 kilometers, while dining room sales keep compounding through word of mouth. Packaging cost appears in no budget and decides margins. Between 0.70 and 1.90 USD per order by category, plus the waste of product that travels badly, leaves a ghost kitchen with an effective food cost 3 to 5 points above what the recipe card shows.

Where each model breaks, and what the break costs — in practice

If your signature dish carries an emulsified sauce or anything fried, better packaging will not close that gap. Changing the dish will.