Ghost kitchen menus in 2026: what still works now the shakeout has taken the hype with it

A ghost kitchen is a commercial kitchen that produces food for delivery only: no dining room, no counter, no passing trade. Every order arrives through a screen. That single fact reshapes the menu, because the menu is no longer a document a guest reads while sitting in your room — it is the entire product, the entire service, and the only salesperson you have.

Most writing on this topic is still calibrated to 2021, when venture money treated ghost kitchens as software with a fryer. That thesis broke publicly between 2023 and 2024. This guide covers what actually happened, what the platforms now require, and how to build a delivery-only menu that survives both the packaging and the commission.

Five terms that get used interchangeably and should not be

The vocabulary confusion here causes real business mistakes, because it merges a commercial real estate business with a menu licensing business.

Term	What it is	Whose brand	Dining room	Where the kitchen sits
Ghost kitchen	Commercial kitchen space built for delivery, often leased to several operators	Multiple tenants	None	Standalone industrial or commercial building
Dark kitchen	Structural synonym for ghost kitchen, more common in Europe and Asia	Multiple tenants	None	Same
Cloud kitchen	Another synonym, originally tied to one company’s naming, now generic	Multiple tenants	None	Same
Commissary kitchen	Licensed shared commercial kitchen, traditionally used by food trucks, caterers and bakers	Multiple tenants	None	Shared commercial building
Virtual restaurant	An existing restaurant that adds a separate delivery-only brand from its own kitchen	The existing restaurant	Yes, for the parent brand	The restaurant’s own kitchen
Virtual brand	A delivery-only brand as intellectual property, produced under licence in partner kitchens	A third party or brand owner	None	Partner kitchens, ghost or traditional

The dividing line is infrastructure versus intellectual property. A ghost kitchen operator is a landlord who happens to supply hoods and grease traps. A virtual brand owner is a tenant or licensee who owns a menu and a logo. That distinction explains why the collapse of the big facility networks did not kill virtual brands: the brands simply moved into other kitchens, including traditional restaurants with idle capacity. For where these formats sit alongside everything else, see our breakdown of the types of restaurants.

What actually happened to the segment

The honest 2026 position: the delivery-only menu is a working format, and the delivery-only real estate bet largely was not.

Venture funding in 2020 and 2021 priced ghost kitchens as if they would scale like SaaS. They do not. The build-out is capital-heavy — ventilation, electrical capacity, plumbing, health code compliance — and the underlying economics were brutal: a restaurant operating at 10% to 15% net margin cannot pay a marketplace up to 30% of every order and grow into profit without continuous external subsidy.

The reckoning is documented, not speculative:

• Kitchen United raised $100 million in 2022 with Kroger among the investors, then closed all eight of its in-Kroger locations — 44% of its 18-unit footprint — and announced it would sell or close every physical site and pivot to software (Restaurant Dive, 2023).
• Reef Technology cut its Wendy’s ghost kitchen partnership from a planned 700 units to a fraction of that, after mass layoffs and site closures.
• CloudKitchens, valued at around $15 billion in 2021, faced occupancy problems and shut acquired properties.
• Grubhub was bought by Wonder for $650 million (Restaurant Dive, November 2024) — the same asset Just Eat Takeaway had acquired for roughly $7 billion in 2020.

The virtual brand layer got cleaned up too, and by the platforms rather than by the market. Uber Eats had removed 8,000 virtual restaurants by July 2023 (Nation’s Restaurant News), roughly 13% of its North American virtual brands, after operators flooded search results with near-identical menus under different names. One San Francisco kitchen was reported running more than 70 storefronts.

What survived is unglamorous and workable: existing kitchens adding one or two genuinely differentiated delivery brands to fill idle capacity, and operators who treat marketplaces as an acquisition cost rather than a business model. If you are weighing the risk of the format generally, our restaurant failure rate statistics hub is the corrective to the folklore in this area: measured first-year failure is 17% to 26%, not the 90% the internet keeps repeating.

Market size figures: read the definition first

Published US market estimates for ghost kitchens range from under $3 billion to over $70 billion, which tells you they are measuring different things. Some firms count commercial kitchen equipment spending, some count the total order value flowing through ghost kitchens, and some count only lease revenue on the facilities. Always take the range with the firm’s name, the year and the definition attached, and never quote a single headline number as “the market.”

How a delivery-only menu differs from a dine-in menu

The core difference is that nothing in the guest’s experience is under your control after the bag is sealed. No server explains a dish, no plating carries the presentation, no room sets expectations. The menu listing, the photo, the description and the packaging are the whole product.

Five practical consequences:

The food has 20 to 40 minutes in a closed box. That box vibrates, cools unevenly and fills with steam. Condensation on the lid drips back onto whatever is below it. This is why fried items that are excellent at the pass arrive soft, and why sauces served over crisp components arrive as one texture.

Photography does the entire job of appetite appeal. DoorDash reports, from a study of over 15,000 small business merchants, that menus with item photos see up to 44% higher monthly sales and menus with item descriptions up to 18% more. That is platform data from a company that benefits from better-looking menus, so treat the magnitude as directional. The direction itself is not in doubt: without a photo, a delivery item is a line of text competing with a photographed one.

The description replaces the server. It has to cover texture, heat level, portion size, what it comes with, and allergens — because there is nobody to ask. Vague descriptions do not create mystique in delivery, they create refund requests.

Modifiers replace the conversation. “Ask your server about substitutions” has no meaning here. Every reasonable variation has to exist as a structured option, or the guest orders somebody else’s dish.

Menu size has to come down. More items means longer prep, more inventory, more assembly steps and more mis-picks, and every mis-pick in delivery is discovered by the customer rather than by staff. The floor is set by the platforms — DoorDash requires a virtual brand menu to carry at least 8 distinct food items, at least half of them hot or prepared — but the working range for most single-kitchen delivery brands sits well under a dine-in menu’s length.

Which dishes survive delivery and which do not

Travels well	Why	Travels badly	Why
Braises, stews, curries	Sauce-based, improve while held hot	Anything battered and fried	Steam condenses inside the box and softens the crust within minutes
Grain and rice bowls	Structurally stable, tolerate temperature drift	Crisp-topped baked dishes	The topping absorbs moisture from what is underneath
Pizza with a sturdy base	Designed for a vented box	Delicate leaf salads with hot dressing	Wilt on contact and keep wilting
Roasted proteins with sauce on the side	Nothing is sitting in liquid during transit	Soufflés, tempura, anything time-critical	The eating window closes before the driver arrives
Pasta held slightly under	Finishes cooking in residual heat	Ice cream and delicate desserts	Need a cold chain the courier does not have
Dumplings, wraps, handhelds	Self-contained, structurally strong	Rare steak and precisely cooked fish	Carryover cooking overshoots during transit

Two rules do most of the work here. Separate the wet from the crisp — sauces, dressings and toppings travel in their own container and get combined by the guest. And vent the steam — a fully sealed container guarantees condensation, so packaging with controlled venting is not an upsell, it is a recipe component.

Budget for it accordingly. Technical packaging for a delivery-only operation typically adds around $1 to $2 per order, which is 4% to 7% of revenue on a mid-sized ticket. Under-spending here is a false economy, because a leaked or soggy order costs the refund plus the rating plus the customer.

How to structure a delivery menu

Order categories by how people decide, not by how kitchens are organised. A dine-in menu can afford starters, mains and sides in classical order. A delivery menu is scrolled on a phone with one thumb, so the first screen has to carry the items you actually want to sell.

A working structure:

1. Signature items first — three to five dishes the brand is for. If the brand is a chicken sandwich brand, the sandwiches are the first thing on screen.
2. Combos and meal deals second — the single most effective ticket-size lever in delivery, because they remove a decision rather than adding an upsell.
3. Sides, sized and priced to be added without thought.
4. Drinks and desserts last, kept short.

Give every item a photo shot to the platform’s specification. Uber Eats and DoorDash both reject images with multiple dishes in frame, overlaid text, logos or promotional banners, and typically want at least 1200 x 800 pixels. Natural, warm lighting outperforms cold artificial light, which makes food look grey.

And keep prep time honest in the platform’s estimate. A menu full of 25-minute items on a platform promising 30-minute delivery produces late orders, and lateness feeds directly into the algorithmic ranking that decides whether anyone sees you at all.

Running several virtual brands from one kitchen

The 2021 version of this strategy was to launch ten near-identical brands from one kitchen to monopolise search results. That is no longer permitted, and the rules are public.

DoorDash publishes virtual brand quality requirements: no more than 10 virtual brands per store address (exceptions for established commissary operators or with written approval); at least 50% differentiation in main menu items from every other menu at that address; at least 8 distinct food items; and at least half the menu hot or prepared food.

Uber Eats certification requires the virtual brand’s menu to be at least 60% different, and both the virtual brand and the parent restaurant to hold a rating of 4.3 or higher, with limits on how frequently new brands can be created.

The operational cost is the part nobody warns you about. Multiple brands on one line means multiple bag types, sticker sets, sauce SKUs and recipe cards at a single station, and assembly errors scale faster than revenue does. There is also review contamination: a bad experience with one brand travels to the operator’s other brands once a customer works out they share a kitchen, and the platforms now display the producing kitchen’s address precisely so customers can work that out.

The version that works: two or three genuinely different concepts that share equipment but not menus — say a grill concept and a rice-bowl concept off the same line — rather than five names for one burger.

The margin math on a $30 order

Here is where the format lives or dies. Take a $30 order at 30% food cost and 5% packaging, and compare a mid-tier marketplace plan against a direct channel.

Line	Marketplace, 25% plan	Direct ordering
Order value	$30.00	$30.00
Food cost, 30%	−$9.00	−$9.00
Packaging, 5%	−$1.50	−$1.50
Marketplace commission	−$7.50	$0.00
Card processing (2.9% + $0.30)	bundled into commission	−$1.17
Contribution	$12.00 (40.0%)	$18.33 (61.1%)

That $6.33 gap per order is not a rounding difference. It is roughly the whole net margin of a healthy restaurant, and it has to cover kitchen rent, labour, insurance, tax and refunds before anything reaches the owner.

Current published marketplace rates run 15% to 30% on delivery depending on plan: DoorDash at 15/25/30% with 6% pickup, Uber Eats at 20/25/30% after its March 2026 increase, and Grubhub splitting 5/15/20% marketing plus a 10% delivery fee and 3.05% + $0.30 processing on top. Our food delivery commission rates hub tracks these against city fee caps and the effective take rate once ads and promotions are included, which lands well above the headline number.

Two consequences fall out of this table. First, delivery-only operators price above their base menu on marketplaces — commonly 20% to 30% higher — because absorbing a quarter of revenue is not survivable. Second, the marketplace is best understood as paid customer acquisition rather than a sales channel: expensive to use, excellent at reaching people who have never heard of you, and something to graduate customers out of. The costing method behind setting both price levels is in our guide to pricing a restaurant menu.

Does a direct ordering channel make sense with no dining room?

Yes, and more than it does for a restaurant with a room, because a ghost kitchen has no other way to own a customer. A restaurant with a dining room collects regulars through the door. A delivery-only brand collects them only if it deliberately builds the path.

What actually moves people across:

• An insert in every bag with a QR code and a first-order discount that beats what the marketplace ever shows them. This is the highest-conversion tool available, because it reaches someone at the moment they are enjoying the food.
• A menu link that works without an app, since asking a customer to install something to save you 25% fails immediately.
• A messaging channel for repeat orders. Reordering a usual through a chat thread has far less friction than navigating a marketplace; our WhatsApp ordering statistics hub covers where that works and where it does not.
• Pricing that is visibly better direct, stated plainly. If the direct price matches the inflated marketplace price, there is no reason to switch.

The realistic target is not to abandon marketplaces. Operators who are still standing typically run a meaningful minority of digital sales through direct channels while keeping marketplace presence for discovery. Our online ordering statistics hub covers how that mix has shifted.

Building the direct channel is mostly a menu problem rather than a technology problem: the same menu data has to render as a page a customer can order from, with photos and modifiers already structured. That is the same work described in how to digitize a restaurant menu, just with delivery constraints applied to the item list.

Regulatory basics

Ghost kitchens and commissary kitchens are regulated as full commercial food production facilities under local, county and state health authorities, working from the FDA Food Code as the model their rules adopt.

The practical requirements:

• A licensed commercial kitchen is mandatory. A delivery-only brand cannot operate from a residential kitchen. Cottage food allowances do not extend to restaurant-style delivery operations.
• The facility needs commercial infrastructure: proper exhaust ventilation, grease interception, a multi-compartment warewashing sink, a dedicated handwashing sink and temperature-controlled storage.
• The permit belongs to the operator, not only the landlord. Renting space in a ghost kitchen does not transfer somebody else’s food permit to your brand.
• Location disclosure is now platform policy rather than federal law. There is no single US statute requiring a virtual brand to publish its kitchen address, but DoorDash and Uber Eats both label virtual brands and display the producing kitchen’s address in the listing, so assume your address is public.

Nothing here is legal advice, and requirements vary by jurisdiction. Check with the health department that will actually inspect you before signing a lease.

A build order that reflects what 2026 rewards

1. Pick one concept, not five. Platform differentiation rules make the shotgun approach non-viable, and one good brand out-earns five clones.
2. Design the menu against the container, not against the plate. If a dish cannot survive 30 minutes in a box, it does not go on the menu regardless of how well it sells in the room.
3. Cap the item count somewhere above the 8-item platform floor and well below a dine-in menu, and defend it.
4. Photograph everything to platform specification before launch, not after.
5. Write descriptions that answer the questions a server would, including allergens.
6. Price two levels — a marketplace price that survives commission, and a direct price that is visibly better.
7. Put a QR insert in every bag from the first order.
8. Measure contribution per order per channel, not gross sales. Gross sales on a marketplace can rise while contribution falls.

For step six and seven, the setup is small: Duckhub’s free Egg plan covers a full menu with photos, modifiers and QR placements at 0% commission on orders, which is enough to run the direct channel that the bag insert points at. The economics above are the argument, not the software — whatever tool you use, the order that arrives through your own channel is worth about half again as much as the one that arrives through somebody else’s.