A ghost kitchen is a commercial kitchen that makes food for delivery and pickup only, with no dining room and no walk-in customers. In 2019 the format was forecast as a trillion-dollar global opportunity by 2030. By 2025 the US market was $2.9 billion and shrinking, and the biggest names in the category had closed, merged, or pivoted to software.
That gap between the forecast and the outcome is the most useful thing about ghost kitchens for anyone choosing restaurant real estate. The model did not fail because delivery demand disappeared. Delivery kept growing. It failed because a delivery-only kitchen inverts almost every assumption in traditional restaurant site selection, and most operators kept using the old assumptions anyway.
What is a ghost kitchen?
A ghost kitchen produces food exclusively for off-premise consumption. No dining room, no drive-thru, no signage that matters, no walk-in trade. Also called a dark kitchen, cloud kitchen, or virtual kitchen, it comes in three formats that get lumped together and should not be.
Shared commissary stall. You rent a kitchen bay inside a facility run by an operator like CloudKitchens or Nimbus, alongside a dozen other brands. Low capital outlay, fast to open, and you are a tenant with a monthly cost rather than an owner with an asset.
Standalone dark kitchen. One brand leases and builds out a space of its own, sited for delivery coverage rather than visibility. This is the format that most resembles conventional retail real estate, and the one where site selection matters most.
Virtual brand inside an existing kitchen. A restaurant with spare capacity launches a delivery-only concept out of its own kitchen. Nothing new gets leased. The real estate decision was already made, so the question is menu and marketplace strategy, not location.
The three formats fail in different ways, and conflating them is part of why the category's numbers are hard to read. A virtual brand launched from an existing kitchen risks a menu. A standalone dark kitchen risks a lease.
What the market data actually says
The trillion-dollar number came from Euromonitor International in November 2019, framed by the firm's Michael Schaefer as a total addressable opportunity for ghost-kitchen-supplied food service by 2030. It described a ceiling for delivery taking share across drive-thru, takeaway, dine-in, and packaged food. It was reported everywhere as the size of the ghost kitchen market. CNBC published a retrospective in February 2024 under the headline "How 'ghost kitchens' went from $1 trillion hype to a struggling business model."
Here is what the measured numbers look like. IBISWorld's 2025 industry analysis puts the US ghost kitchen market at $2.9 billion, with revenue down 5.2% in 2024 and a negative compound growth rate across 2020 to 2024, even as the count of businesses in the category kept edging up. More operators, less revenue each.
The forward-looking forecasts have not gotten more disciplined. Coherent Market Insights put the 2030 global market at $157 billion in a July 2023 report. A Research and Markets report published in July 2026 put it at $204 billion. Other trackers covering roughly the same category land lower still.
A $47 billion gap between two forecasts of the same market in the same year, from firms selling the same kind of report, is not a rounding difference. The category has no agreed method for sizing itself. Treat any headline ghost kitchen market number as a marketing artifact and work from the measured revenue instead.
The shakeout, 2022 to 2024
The category's contraction is well documented and specific.
Reef Technology paused operations at 95 underperforming ghost kitchens in January 2022, roughly a third of its footprint, then cut about 750 jobs in May 2022. Wendy's, which announced in 2021 that it would open as many as 700 Reef locations by 2025, trimmed that target to 100 to 150 units, then closed its remaining roughly 15 US Reef units in the second quarter of 2023 and exited the partnership.
Kitchen United closed all eight of its in-store Kroger locations in November 2023, roughly 16 months after Kroger joined a $100 million investment round in the company. In February 2024 it shut its remaining physical locations and pivoted to licensing software. Sam Nazarian acquired the leftover assets in March 2024 and folded them, along with Nextbite and C3, into a single entity.
Nextbite went through several rounds of layoffs across 2023 and sold its Ordermark US business to UrbanPiper before the Nazarian acquisition. CloudKitchens, valued at $15 billion after a roughly $850 million raise in late 2021, was reported by the Financial Times in September 2023 to be closing warehouses and cutting staff, with occupancy in its buildings around 50%.
By March 2024 the virtual brand business had consolidated down to two significant independent players. Nazarian, who bought most of the wreckage, was blunt about what he had bought into: "I think fundamentally, if you're a delivery-only business, it's not a profitable endeavor."
Why the unit economics broke
Alon Lagstein, counsel at the law firm Carlton Fields, gave Restaurant Dive the cleanest diagnosis in May 2024: "The pandemic gave the business model a false positive." Demand spiked, capital followed, and operators expanded before the math worked. Three specific problems sat underneath.
Marketplace commissions take a fixed cut of a delivery-only business's entire revenue. DoorDash's published merchant rates are 15%, 25%, and 30% for delivery orders depending on plan. Uber Eats publishes 20%, 25%, and 30%, with an additional 5% on Uber One orders for merchants on the Plus plan. A restaurant with a dining room absorbs those rates on one channel. A delivery-only kitchen pays them on everything it sells.
The real estate advantage was temporary. Shared ghost kitchens sold cheap entry against the cost of a full restaurant buildout. Then the 2020 to 2022 closure wave flooded the market with second-generation restaurant space carrying usable kitchen infrastructure, and the arbitrage closed. Lagstein names this directly: cheap conventional space erased the reason to pay a premium for a commissary stall. Anyone comparing a commissary stall to a second-generation lease today should run both against current retail buildout costs rather than against the 2021 version of that comparison.
The platforms rationed supply. Virtual brands on Uber Eats grew from about 10,000 in 2021 to more than 40,000 by March 2023, including cases of a dozen brands selling the same food from one kitchen. Uber Eats announced a cleanup in March 2023 and had removed 8,000 virtual brands by July. Its policy requires a virtual brand's menu to differ from the parent restaurant and any co-located brands by roughly half its items, requires photos of five unique menu items, and removes brands averaging below 4.3 stars, a higher bar than physical restaurants face. DoorDash's published virtual brand quality requirements run parallel, at 50% menu differentiation between co-located brands. The play of spinning up ten storefronts from one kitchen to buy shelf space is closed.
Meanwhile the demand story stayed healthy, which is what makes the collapse instructive. The National Restaurant Association reported in April 2025 that nearly 75% of restaurant traffic is now off-premise. DoorDash reported 2025 orders up 23% to 3.2 billion. Delivery is not the problem. Delivery-only real estate was.
What actually changes about site selection
Strip out the failures and the real question is still there: how do you pick a site for a kitchen that no customer will ever see? Nearly every input in a conventional restaurant site model either flips or drops out.
Visibility becomes worthless. Signage, corner position, and traffic counts on the adjacent road are the backbone of quick-service site scoring. For a dark kitchen they are noise you are paying rent for. The site is discovered inside an app, not from the road.
The trade area compresses. A sit-down restaurant may draw from five to seven miles. DoorDash sets the delivery radius dynamically, based on demand, courier availability, and partnership tier, and does not publish a fixed number. Third-party estimates put the typical radius around 5 miles, tighter in dense markets and wider in rural ones. Many operators plan closer to 3 because food quality falls off with drive time. That is a trade area with roughly a third the land area of a conventional one, so the density inside it has to carry the whole forecast. A shrinking radius makes the trade area saturation question harder, not easier.
Population density replaces foot traffic. With no walk-in trade, the variables that matter are how many households and daytime workers sit inside the delivery radius, what they order, and at what hours. This is why foot traffic data providers are the wrong first purchase for a delivery-only format. Visit counts measure the exact behavior your site will never produce.
Driver supply becomes a site variable. Order throughput depends on couriers being available near the kitchen at peak. Platforms adjust radius by driver density, so a site in a courier-thin area gets a smaller effective catchment than the map suggests. This one has no clean public dataset behind it, which is precisely why operators find out after signing.
Zoning opens up and narrows at the same time. A kitchen with no public dining occupancy can sometimes sit in industrial or mixed-use zoning under lighter permitting than a full restaurant, and several cities have written dedicated districts for delivery-only food production. That widens the search. It also means the comparable-rent set is industrial flex space rather than retail, and the vetting list adds ventilation capacity, utility service, waste handling, and driver ingress and staging. Delivery-only kitchens sit in the same real estate category shift as dark stores and micro-fulfillment: retail demand served from industrial space.
How to evaluate a delivery-only site
The method that survives the shakeout is unglamorous. Model the demand inside a tight radius, price the channel honestly, and check the site against the rest of your portfolio before signing.
- Draw the real radius, not the aspirational one. Start at 3 miles and only widen it if you have evidence of courier density and order data supporting the longer drive. Everything else in the model depends on this boundary.
- Count households and daytime population inside it, by daypart. A dense office core and a dense residential core produce different order curves from the same headcount. A dinner-weighted concept in a district that empties at 6 p.m. has a demand problem no marketing budget fixes.
- Count the competing supply inside the same radius. Marketplace competition is the shelf you are fighting for. Category density inside 3 miles matters more than the tenant next door.
- Price the channel before the rent. Model 15% to 30% marketplace commission on the full revenue line, not on a delivery slice. If the site only works at the lowest commission tier, it does not work.
- Check overlap with your existing locations. A dark kitchen placed to backfill delivery coverage can quietly eat orders from a nearby store you already own, which is cannibalization with none of the offsetting walk-in trade to soften it.
- Verify zoning, ventilation, utilities, and driver access before the LOI. These are the items that kill delivery-only deals late, after the demand case is already approved.
- Compare the total cost against a second-generation restaurant lease. The commissary stall may still win on speed and capital. Since 2023 it has stopped automatically winning on cost.
That sequence is the same site selection discipline any expansion team runs, with the inputs reweighted. What went wrong in the ghost kitchen boom was not the absence of a method. It was applying a method built for visibility to a format that has none.
Where the analysis has to be honest
The unglamorous conclusion in the data is that delivery-only real estate works as one channel inside a business, not as the whole business. Wonder is the closest thing the category has to a winner, and it got there by acquiring Blue Apron and Grubhub and running physical food halls, not by staying delivery-only. The operators still opening dark kitchens successfully in 2026 tend to be established brands adding delivery coverage in a market they already understand, where the kitchen extends a demand model that is already validated by real stores.
That is a forecasting problem before it is a real estate problem, and it is the part most of the 2021 cohort skipped. Modeling demand for a new site means using the trade-area variables that predict orders, not the ones that predict walk-ins, and being able to explain which inputs moved the number. GrowthFactor scores sites on trade-area demographics, competitor proximity, and overlap with your existing locations, and shows the inputs behind the score so your team can inspect the logic rather than accept a number. For a delivery-only format, the site scoring lenses get reweighted toward density and away from traffic counts, and the deal pipeline holds the zoning and buildout diligence next to the score instead of in a separate spreadsheet.
The ghost kitchen boom produced a lot of leases signed against a forecast nobody could inspect. Making the demand case legible before the lease is the entire lesson.
Frequently Asked Questions about Ghost Kitchens
Here are concise answers to common questions about ghost kitchens and delivery-only site selection.
What is a ghost kitchen?
A ghost kitchen is a commercial kitchen that produces food for delivery and pickup only, with no dining room and no walk-in trade. It takes three common forms: a rented stall inside a shared commissary, a standalone dark kitchen leased and operated by one brand, and a virtual brand run out of an existing restaurant's kitchen during slow hours.
Are ghost kitchens still profitable in 2026?
Some are, but the category as a whole shrank. IBISWorld put the US ghost kitchen market at $2.9 billion in 2025, with revenue down 5.2% in 2024 and a negative growth rate since 2020. The operators still making money tend to run a delivery-only kitchen as one channel inside a larger business, not as the whole business.
How big should a ghost kitchen delivery radius be?
DoorDash sets the radius dynamically based on demand and courier availability rather than publishing a fixed default, and third-party estimates put the typical radius around 5 miles, tighter in dense urban markets and wider in rural ones. Many operators plan closer to 3 miles because food quality falls off with drive time, which makes population density inside that tight radius the dominant site variable.
Why did so many ghost kitchen companies fail?
Three things stacked. Delivery marketplace commissions run 15% to 30% of the order before add-on fees. The pandemic wave of restaurant closures flooded the market with cheap second-generation restaurant space, erasing the cheap-sublease advantage that made shared kitchens attractive. And the delivery platforms cracked down on duplicate virtual brands, with Uber Eats alone removing 8,000 of them by July 2023.
How does GrowthFactor compare to Placer.ai for delivery-only site selection?
Placer.ai measures foot traffic, which is the signal a delivery-only kitchen does not generate. A dark kitchen has no walk-in trade to count, so visit data tells you little about whether the site will work. GrowthFactor is a site selection and deal pipeline platform for retail expansion teams, and it scores sites on the trade-area variables that do drive delivery demand, including residential and daytime population density, competitor saturation inside the delivery radius, and overlap with your existing locations.