Physical restaurant vs dark kitchen: the prices nobody puts in writing

What does each model cost to open as of September 2026

Opening a physical restaurant of 80 to 120 square meters in a large Latin American city costs between 120,000 and 450,000 USD, while a station in a shared kitchen hub starts at 8,000 to 25,000 USD for setup plus monthly rent of 700 to 2,400 USD. The gap looks overwhelming until you check the second column. In the physical venue, 35% to 55% of the equipment keeps resale value after three years, and the goodwill can be transferred; in the hidden kitchen, you hand back the key and walk out with a fryer. Almost nobody discounts that when comparing the two opening numbers. The physical investment buys ASSETS, the dark kitchen buys access, and access gets paid again every month you want to keep cooking there. Each range buys different things, and it deserves to be said plainly. From 8,000 to 25,000 USD you set up a hub station: shared hood, two or three owned appliances, packaging, listings on two apps and six weeks of working capital; no storefront, no dining room.

What each investment range actually includes?

From 40,000 to 110,000 USD you can build a standalone hidden kitchen in your own unit, with permits, real extraction, cold storage and a twelve-week cushion.

Between 120,000 and 260,000 USD you get a neighborhood restaurant with 40 to 60 seats, light construction, furniture, licenses and three months of payroll covered. From 260,000 to 450,000 USD we are talking about a destination venue: custom build, bar, full cooking line, signage and a floor team trained before opening. The classic mistake is buying the upper range with the working capital of the lower one. The cost that decides this model is the acquisition cost of every order, and inside a dark kitchen it hides within the platform commission. As of September 2026, delivery commissions across the region run between 18% and 30% depending on category, city and whether you accept the app's courier or use your own; geolocated advertising inside that same app, which hardly anyone skips now because without it you never rank on top, adds another 4% to 9% of sales.

The deciding cost is not rent, it is commission

Add those two and you understand why so many virtual projects close with a full kitchen. A physical restaurant pays for customer acquisition differently: a well-maintained Google Business Profile, with exact hours, correct primary category and fresh photos every month, costs the same whether it brings 900 discovery visits or 2,400. The first difference is structural, not a matter of amount: in the hidden kitchen the acquisition cost grows exactly at the pace of your sales, while in the physical venue it gets diluted. Put numbers on it. If you bill 40,000 USD in a month running a dark kitchen, somewhere between 8,800 and 13,600 USD of that revenue left in commission and advertising; bill 80,000 and you paid twice as much, with no economy of scale to rescue you. In the dining room, the work of maintaining the listing, answering reviews and refreshing photos costs the same every month and spreads across more tickets as you grow.

Variable against fixed: the structural difference

Diego F. Parra puts it this way during Masterestaurant audits: a cost that climbs in a straight line with revenue is not marketing, it is an equity stake you handed an app without signing anything. Five factors explain almost all the spread in these ranges. City and corridor come first: a square meter in a premium zone multiplies rent by 2.5 or 3 against the same meter twelve blocks away. Condition of the unit comes next, because building from scratch adds 25% to 40% on top of the equipment budget. Extraction and smoke permits alone can move 15,000 to 60,000 USD, and in older buildings they decide whether the project exists at all. Channel mix drives the recurring cost: every ten points of sales that shift to third-party apps take between 1.8 and 3 points off your operating margin. And the fifth, the one almost nobody budgets, is working capital: twelve weeks of payroll and suppliers, not six.

So which number decides?

The deciding number is the share of your projected revenue that will flow through a third-party app, and the threshold sits near 55%.

Above that line, the dining room becomes a fixed cost you pay to serve a customer who never walks in, and the dark kitchen makes economic sense. Below 40%, the physical venue wins comfortably, because every table served in-house avoids 18 to 30 points of commission and also generates reviews, repeat visits and a higher check. The middle band, between 40% and 55%, is the one that manufactures bankruptcies: you pay dining room rent, app commission and advertising all at once. Before signing any contract, project that percentage with data from your category and your neighborhood, not with the optimism of the business plan. Four levers actually move the needle, and none of them is asking for a discount over the phone.

How to negotiate and cut the real bill?

First, negotiate commission against committed volume: above 15,000 USD in monthly sales on a single app, two or three points are negotiable, and two points on 180,000 USD a year is 3,600 USD back in the till.

Second, run your own courier on short-radius orders, where the platform's logistics fee weighs more than the marketplace commission. Third, negotiate the hub on 24 months with three months of grace instead of accepting the list price at 12. Fourth, the one that pays best, move orders to your own channel: each point you recover is worth about 150 USD clean on 60,000 USD of monthly sales. Open your P&L this week and break commission, advertising and logistics into THREE separate lines. An operator running two virtual brands from a Bogotá hub was billing 62,000 USD a month and believed the business was healthy because food cost sat at 29%.

The case that shows up in every audit

When his P&L was opened with the Masterestaurant team, platform commission took 26%, in-app advertising took 7% and the station cost 1,900 USD; real operating margin landed at 4.1%, against the 11% he assumed. He shifted 30% of his orders to direct WhatsApp ordering with in-store pickup, renegotiated the station on 24 months and lifted margin to 9.3% in five months without selling a single peso more. Nobody had overcharged him: he had simply never seen commission as a line of its own. That diagnosis repeats constantly, and it does not get fixed by switching models, it gets fixed by reading the right cell. The first one is structural rather than a matter of size: in a dark kitchen the acquisition cost is VARIABLE and grows at exactly the pace of your sales, while in a physical unit it is fixed and dilutes. Sell 40,000 USD in a month from a hidden kitchen and you paid 8,800 to 13,600 USD for those sales in commission and ads; sell 80,000 and you paid twice as much.

Three price differences that decide it

In the physical model, the Google Business Profile listing that brought 900 discovery visits cost the same whether it brought 900 or 2,400, because what you pay for is the upkeep, not each click. Second comes what the startup capital actually buys. Those 120,000 to 450,000 USD do not evaporate: part stays in equipment worth 35% to 55% of its price at year three, part in goodwill that transfers with the lease, part in construction that is genuinely gone. Inside a dark kitchen, of the 18,000 to 55,000 USD, the recoverable slice rarely clears 40%, since most of the capital goes into branding, product photography, packaging, working capital and the hub's security deposit —money that never comes back if the project goes dark. The third one is hardest to see and it belongs to the algorithm.

Three price differences that decide it — in practice

Rappi, Uber Eats and DiDi Food rank by recent conversion, prep time and rating, and they punish cancellations and delays; a virtual restaurant with no room depends 100% on that borrowed shelf, so one weighting change —the kind platforms ship two or three times a year without warning— can cut 30% of your revenue in a fortnight. A physical unit also rides Maps algorithm shifts, yet it keeps the door, the sign, and the guest who already knows the way.