AI girlfriend app market size 2025, 2026, and 2027

The AI girlfriend is not a chatbot with a pretty skin. She is a named companion with a face, a memory, and a monthly charge. People come back after midnight. They pay to keep her.

Analysts now treat that aisle as a real market. The Business Research Company put it at $2.32 billion in 2025. Research and Markets had $2.91 billion in 2026. Push the same growth rates one more year and 2027 lands in a $3.5–$5.2 billion band.

That is the forecast. The till is smaller. Appfigures saw $162.8 million spent on romantic and NSFW companion apps in the first half of 2026, on the App Store and Google Play. Web girlfriends such as Candy AI sit outside that till. Both numbers are true. They are not the same number.

This is not the dating-copilot story. Bios, pickup lines, and first-date scripts live in AI relationship statistics 2026. This piece is the cash. 2025. 2026. 2027. Country. Genre. The apps that take the money.

The aisle
What an AI girlfriend app actually sells

An AI girlfriend app sells a relationship-shaped product. Chat that remembers. A face that can send a still. Often a voice. Sometimes a short video. A name that stays the same next week.

That is a different shelf from a workplace bot. It is a different shelf from ChatGPT rewriting a Tinder bio. Grand View Research puts a wide “AI companion” TAM at $36.8 billion in 2025 and $48.0 billion in 2026. That TAM eats device assistants and office tools. Quote it as “AI girlfriends” and the figure lies.

XXXFuel scores the girlfriend aisle. Named companions. Memory. A photo that belongs to the thread. Candy, OurDream, GirlfriendGPT, Kindroid, Nomi. Libraries such as Character.AI sit next door. They share users. They do not share a TAM.

The bill is sticky. Sessions run long. Tokens top up the stills. Annual plans lock the habit. That is why this aisle compounds even when the wider chatbot market cools.

2025
The year the aisle got a price tag

2025 is the first year the desks agree there is a market, even when they disagree on the size.

The Business Research Company printed $2.32 billion for 2025. Software took 74.6% of that. Personal companionship took 48.1% — the largest use, and the fastest. Freemium took 49.3%. Subscription is the clip they say will outrun it, at 27.2% CAGR through 2030.

SNS Insider printed a fatter 2025: $3.08 billion. Android led platform share at 32.33%. North America led region at 40.07% on that desk. The U.S. slice alone was $0.89 billion.

Dataintelo sat in the middle at $2.8 billion for 2025, heading to $11.6 billion by 2034 at 17.1% CAGR — the slowest of the girlfriend-app forecasts.

On the phone till, 2025 was smaller and sharper. TechCrunch, citing Appfigures in August 2025, said companion apps were on track for more than $120 million that year. 220 million lifetime downloads by July. 60 million downloads in the first half, up 88% year on year. Revenue per download jumped from $0.52 to $1.18.

128 of 337 live companion apps had launched in 2025 already. The top tenth took 89% of the spend. 17% of the apps put “girlfriend” in the name. 4% said “boyfriend”. The demand is not evenly split. XXXFuel’s own search split says the same thing: AI boyfriend searches are a thin slice of girlfriend demand.

2026
The year the spend compounded

2026 is the year the forecasts and the receipts both moved. They still do not meet.

Research and Markets is the number XXXFuel already uses on the desk: about $2.91 billion for the girlfriend-app market in 2026. TBRC’s 22.2% clip from a $2.32 billion 2025 lands near $2.84 billion. SNS’s 20% clip from $3.08 billion lands near $3.70 billion. Market.us, running a hotter 24.7% from a $2.7 billion 2024 base, prints about $4.2 billion.

Call 2026 a $2.8–$4.2 billion forecast cluster. The centre of gravity is still the low threes.

The till told a louder story. Decrypt, with Appfigures Intelligence in July 2026, counted $427.3 million of all-time consumer spend on romantic and NSFW companion apps since late 2022. 165.3 million downloads. 214 apps.

Thirty-eight per cent of that all-time total landed in the first six months of 2026. $162.8 million on romantic and NSFW apps. Another $164.8 million on general-purpose companions. About $328 million in one half-year, on storefronts alone.

That is compounding. It is still not $2.9 billion. The gap is the point of the next chart.

Who took the 2026 storefront cash? In the romantic and NSFW bucket, Zeta led the first half at $33 million. Tipsy Chat followed at $15.2 million. Then ChatBox at $13 million, Crushie AI at $8.8 million, and Emochi at $7.5 million. Emochi also led installs, at 7.9 million.

Crushie is on XXXFuel as Crushi. Candy is not on that list. Candy bills on the web. Apple never sees the card. That is a feature, not a hole in the product.

2027
The year already in the pipeline

No desk has printed a clean “2027 TAM” as a headline. The 2027 number is a clip, not a prophecy. XXXFuel still runs it. Plans for 2027 are being built on 2026 cash.

Hold each desk’s CAGR still for one more year:

• TBRC — about $3.46 billion in 2027 (22.2% off $2.84B).
• Research and Markets — about $3.65 billion if the 2025→2026 jump repeats.
• SNS Insider — about $4.44 billion (20% off $3.70B).
• Market.us — about $5.24 billion (24.7% off $4.20B).

So 2027 is a $3.5–$5.2 billion band. The honest centre is the high threes to low fours. Anyone selling “$10 billion by 2027” is spending a later year early.

Further out, the desks fan apart. TBRC sees $6.32 billion in 2030 and $16.4 billion in 2035. SNS sees $11.06 billion by 2032 on the older 20% path, and $19.09 billion by 2035 on the later print. Market.us runs to $24.5 billion by 2034. Those tails assume the habit holds. Habits can break. Stores can ban. Banks can drop the merchant code. The 2027 band is the one a builder can actually plan on.

The gap
Why $163 million and $2.9 billion can both be true

Clone sites love one fat TAM. They paste $25 billion next to Candy’s logo and call it a market. XXXFuel does not.

Three books sit on this desk.

Book one: storefront receipts. Appfigures. Apple and Google. Visible. Incomplete. Web-only girlfriends vanish. Candy AI bills on the open web, on purpose. So do GirlfriendGPT and a long tail of NSFW companions. A store ban is not a business model they want.

Book two: girlfriend-app TAM. TBRC, SNS, Research and Markets, Market.us. They try to count subscriptions, tokens, and web cash. They also count hope. That is why 2025 already spans $2.32–$3.08 billion.

Book three: “AI companion” TAM. Grand View. $36.8 billion in 2025. $48.0 billion in 2026. $318 billion by 2033 at 31% CAGR. Social companionship is only 29.4% of that 2025 mix. Mental-health bots and device assistants fill the rest. Useful. Not a girlfriend.

The honest sentence is short. People are spending hundreds of millions, on the stores, right now. The aisle is bigger than the stores. It is not $48 billion of girlfriends.

Country
North America holds the till. Asia holds the clip.

North America took 33.4% of TBRC’s 2025 market — about $775 million. SNS put the same region at 40.07%. The U.S. slice on SNS was $0.89 billion in 2025, on a path to $6.61 billion by 2035 at 22.19% CAGR. An older SNS print had the U.S. at $0.73 billion in 2024 heading to $2.98 billion by 2032.

That is the money. It is not the whole map.

Asia-Pacific is the fastest clip in the same reports. India is the country Grand View flags for companion-app speed. China is a different product-market. Wired (January 2026) described Gen Z women there dating AI boyfriends on MiniMax’s Xingye and on Zhumengdao, with Tencent and Baidu in the aisle. Outside China, a web-analytics pass on the top 55 companion platforms still ran about eight men to two women.

Europe sits between those poles. English-language NSFW girlfriends travel. Payment rails do not. A UK or EU card can die on an adult merchant code that a U.S. processor still clears. Crypto exists for that reason. XXXFuel lists it when the vendor actually takes it.

Country is also taste. U.S. traffic leans realistic girlfriends and uncensored chat. Japan and Korea carry a heavier anime and bunny-girl share. Brazil and Spanish-speaking markets punch above their TAM weight on web girlfriends. The stack is the same. The girl is not.

Genre
Bunny, country, fantasy — the look is the product

People do not buy “an LLM”. They buy a look that answers. The 2026 catalogue is a wardrobe, not a model card.

Realistic is still the cash aisle in the West. A face that could sit in a dating app. Silk. Warm light. Candy, OurDream, and a long web tail sell this first. It converts because it feels like a person, not a costume.

Anime and bunny are the other engine. Bunny ears. Waifu faces. A body the real world will not ship. This is not a side quest. It is a large share of image tokens and a loyal annual plan. Japan-facing traffic treats it as the default, not a fetish tag.

Country is the American flavour of the same trick. Hats. Dust. A girl who looks like a weekend, not a city. It photographs well. It sells on web catalogues that let a user build a look in two minutes. It is smaller than anime. It is stickier than a generic brunette.

Fantasy — elves, mages, royal silk — sits with roleplay. This is where Character.AI, Janitor AI, and CrushOn still win on volume, even when the girlfriend apps win on heat. A named elf with inventory is a different job from a girlfriend who texts good morning.

TBRC does not print a bunny-versus-country split. The desk still sees the mix in galleries. A 2026 companion that cannot do at least two of these looks leaves money on the table. Image tokens are the proof. Chat keeps them. Stills upsell them.

Who takes the cash
Web girlfriends versus storefront girls

The 2026 aisle is two checkouts.

Storefront girls live on Apple and Google. Zeta, Tipsy, Crushie, Emochi. They pay the store tax. They get distribution. They show up in Appfigures. A policy change can kill a year of growth in a week.

Web girlfriends live on a URL. Candy. OurDream. GirlfriendGPT. Soulkyn. Kindroid. They take cards, and often crypto. They keep the cut. They vanish from every “app revenue” chart that clone blogs then cite as the whole market.

Candy is the web case the aisle still copies. Annual Premium has sat near $3.99 a month on a year’s bill, with a steeper monthly. Tokens sit on top. Voice and Shorts are the upsell, not the hook. Clone sites throw around $25 million to $50 million ARR. Candy does not publish. XXXFuel does not pretend those screenshots are audited. The product is real. The audited ARR is not on this desk.

Five companions XXXFuel actually opens when the question is “girlfriend, not generator”:

Libraries still matter for volume. Character.AI is the ceiling on monthly actives — and the legal risk. Kindroid and Nomi sell memory as the product. Best AI girlfriend apps in 2026 groups them by job, not by TAM.

The bill
Subscriptions, tokens, and the annual trap

Almost every serious girlfriend app now runs the same till.

A free taste. A monthly that looks dear. A quarter or a year that looks cheap. Tokens for stills, voice, and video so the subscription never quite covers the hunger.

TBRC says freemium was 49.3% of 2025 revenue. Subscription is the faster clip at 27.2% CAGR through 2030. That matches what the desk sees on vendor pages. Free is the hook. Annual is the lock. Tokens are the margin.

Typical paid bands in 2026:

• Monthly — about $10–$20.
• Annual — about $4–$8 a month, billed as a lump.
• Tokens — stills and shorts on a second metre.

Appfigures’ $1.18 revenue per download in 2025 is a storefront number. Web girlfriends can run far higher ARPU because the heavy user never hits a store wall. They also churn in silence. There is no App Store review to warn the next buyer.

Crypto is not a personality. It is a backup rail when an adult merchant code fails. XXXFuel notes it when the checkout actually offers it — Candy, several NSFW webs, not the Apple list.

2027 product
Video, voice, and a memory that does not drift

The 2027 bill will not come from a fatter TAM slide. It will come from three product moves already in beta on the aisle.

Live video. Shorts are 2026. A face that talks back in a window is 2027. Candy Shorts, FLIRTcam, and a handful of “call” plans are the rehearsal. Compute is the brake. Price is the second brake. The user who pays for a window will not go back to stills.

Voice that stays in character. Voice notes are cheap. A live call that does not snap out of the girl is not. Kindroid and a few memory-first apps already sell this as the reason to stay. 2027 is when it becomes table stakes, not a deluxe toggle.

Memory that does not drift. The complaint is always the same. She forgot the dog. She forgot the anniversary. She forgot the rule. DemandSage’s Character.AI recap still had memory as a top ask. The girlfriend apps that win 2027 will be the ones that keep a private ledger per user, not a pretty reset.

Regulation sits under all three. Age gates. Non-consensual stills. Store rules. Bank rules. A 2027 TAM that ignores those is fan fiction. The adult-content share piece is the adjacent risk map. This piece stays on the girlfriend till.

How to read a TAM
A builder’s cut, not a brochure

Use 2025–2027 like this.

If the job is a pitch deck, the girlfriend-app cluster is $2.3B (2025) → ~$2.9B (2026) → ~$3.5–$5.2B (2027). Cite the desk. Do not average them into a fake precision.

If the job is a build, start from receipts. Storefront companion apps did about $120 million in 2025 and $163 million of romantic/NSFW spend in H1 2026. Web cash sits on top and is opaque. A new app does not get 1% of $2.9 billion. It fights for a sliver of a top-tenth that already takes 89% of storefront spend.

If the job is a girlfriend a person will keep, skip the TAM. Open Candy for the catalogue-plus-shorts bundle. Open OurDream to make one. Open Kindroid when memory is the product. Open the dating category when the question is the whole aisle. How the girl is built is a different article: how to make an AI girlfriend.

XXXFuel will update the 2027 band when a desk prints a real 2027, or when Appfigures drops a full-year 2026. Until then, the band above is the interpolation, labelled as one.

FAQ
Questions people actually type

Citations
Sources

Last updated 5 September 2026. Forecasts move. Vendor prices move. If a homepage slider and a till receipt disagree, the till wins.