The AI Companion App Market: Size, Players and Where the Numbers Go Wrong

If you look up the AI companion app market, you get numbers that can't all be right. One source says $36.8 billion in 2025. Another says $30.1 billion in 2026. A third says $5.01 billion in 2026. A fourth says $764 million in 2024.

They're not measuring the same thing. Some count enterprise meeting assistants. Some count virtual characters in games. Some count only what shows up in app store billing, which misses the web subscriptions where most adult spending actually happens.

This is a piece about the gap. What each estimate includes, what the download and revenue data shows, who's making money, and what could break the growth story. We run an AI companion platform, so we have a stake here — but we also have a reason to be honest about the numbers, because we see what people actually pay.

Market size: four estimates, four different worlds

Here's the table. Every figure is attributed to its source. None of these are ours except the last row.

Source	Measures	Size	Forecast	Growth
Grand View Research	Broad AI companion market	$36.8B (2025)	$318B by 2033	31.0% CAGR 2026–2033
Globe Market Research	AI companion apps	$30.1B (2026)	$198.3B by 2035	n/a
Research and Markets	AI companion apps	$5.01B (2026)	$9.65B by 2030	17.8% CAGR
Research and Markets	AI girlfriend apps	$2.91B (2026)	$7.15B by 2030	25.2% CAGR
SNS Insider	AI girlfriend apps	$2.57B (2024)	n/a	n/a
LinkedIn industry article	AI companion platforms	$764M (2024), $856M (2025)	$1,666M by 2032	n/a
Appfigures via TechCrunch	Mobile spending on AI companions	$82M H1 2025; over $120M 2025	n/a	+64% H1 year over year
Sensor Tower	Mobile spending, broader definition	$150M Q1 2026	n/a	Over 12x Q1 2023
Our estimate	App stores plus web billing	About $1–1.5B a year (2026)	n/a	Low-medium confidence

Look at the spread: $5 billion to $37 billion for roughly the same period. That's a factor of seven. Analyst summaries of this space range from several billion to over $40 billion, which tells you nobody has agreed on what "AI companion" means yet.

Why the estimates disagree

Three things are happening at once.

1. The definition problem

"AI companion" is a fuzzy label. Grand View's broad read includes far more than consumer girlfriend apps — the kind of category that can absorb enterprise assistants, virtual characters, and anything with a conversational persona. Zoom literally calls its meeting assistant AI Companion. It's an enterprise tool, and it is not in the same business as a chatbot you talk to at 1 a.m.

When a report says "AI companion apps," check the scope. If it's a multi-billion-dollar number, it's almost certainly counting something bigger than the apps you'd find on the App Store under that name.

2. Trackers can't see the web

Appfigures and Sensor Tower measure app store billing. That's real, and useful, and it misses a lot. Most adult spending sits in web subscriptions, not in-app purchases, because app stores have rules about sexual content.

Platforms like Candy.ai, SpicyChat, CrushOn, Joi and OurDream AI run largely or entirely on the web. The trackers don't see those subscriptions. Sensor Tower's broader definition implies roughly a $600 million mobile run-rate — real money, but nowhere near the tens of billions some reports claim.

3. Forecasts built on thin data

Sensor Tower's baseline is Q1 2023. That means forecasts out to 2033 or 2035 are extrapolating from about three years of data — and the first of those years was pre-mainstream. Treat any projection past 2030 as a scenario, not a prediction.

The growth that is actually measurable

Setting aside the size fight, the mobile data tells a consistent story. Appfigures reported about 60 million downloads in H1 2025, up 88% year over year. Lifetime downloads globally passed 220 million.

Spending per download climbed from $0.52 in 2024 to $1.18 in 2025. That's the more interesting number. It means people aren't just trying these apps — they're paying more once they're in.

By July 2025, 337 companion apps were earning money, and 128 had launched that year. In H1 2026, companions made up about 4% of over $4 billion in AI app spending.

Analyst CAGRs for the category cluster between 18% and 31% a year. Fast, but not the hockey stick the headline numbers imply.

Who's using these apps

The user base is broader than the stereotype.

• Common Sense Media found 72% of US teens had tried AI companions and 52% were regular users.
• Wheatley and the Institute for Family Studies found 19% of Americans had chatted with an AI romantic partner.
• Match and Kinsey found 16% of US singles had used AI as a romantic companion.
• Broken by gender: 31% of US men aged 18–30 and 23% of women in that range had chatted with an AI partner.
• About a third of teens used companions for social interaction, per Common Sense.

Gender splits get messy depending on the platform. Ada Lovelace found 77% of users on mixed-use UK platforms are male. JanitorAI claims 70–80% women, while Similarweb desktop data shows 37.3% — a gap big enough to make you distrust the company line.

Naming is skewed too. Appfigures found 4% of companion apps named "boyfriend" against 17% named "girlfriend."

On the model side, OpenRouter data shows role-play accounts for roughly 52% of tokens on open-source models. Text-based companions held 42.7% of the market in 2025, per Grand View.

Regions: North America leads, Asia builds

Grand View puts North America at 33.5% of revenue in 2025. Other summaries put it at roughly 35% to 38%. Either way, the US and Canada are the biggest single market.

The UK is a notable secondary hub. Ada Lovelace found Character.AI is a significant UK player, and UK adoption of mating-oriented AI runs 14% above the global average.

Asia Pacific is where the product innovation is coming from. MiniMax listed in Hong Kong in January 2026, raising about US$620 million. Its companion apps were about 35% of $53.4 million in revenue over the first nine months of 2025.

Love and Deepspace passed $750 million, about 60% of it from China, according to KrASIA. Another source puts it at $820 million. The discrepancy is normal for this space.

There are no reliable bottom-up figures for South Korea or the Middle East. If someone quotes you one, ask where it came from.

Key players and what they earn

Company-reported numbers are self-reported unless noted. Treat them accordingly.

Player	Type	Reported scale	Money signal	Watch out for
Character AI	SFW chat	~20M monthly users (Sacra)	~$30M annualized July 2025 (Sacra)	Under-18 ban; five teen-harm suits settled
Chai	18+ chat	10M+ users (company)	$80M ARR Q1 2026; $100M+ claimed July 2026	Self-reported figures
Talkie (MiniMax)	SFW chat	~20M monthly users with Xingye (ChinaTalk)	~$5 per customer over 9 months (ChinaTalk)	Pulled from US App Store December 2024
Replika	Emotional	40M+ users (company)	$24–30M (Sacra earlier) vs $4.8M (Latka 2026)	€5M GDPR fine, Italy
JanitorAI	Role-play	2.5M daily users (company, via Forbes)	Janitor+ $12.99, June 2026	Audience claims conflict
SpicyChat	Role-play	100M registered (company)	Prices raised September 8, 2026	Recent price rise
Nomi AI, Kindroid	Emotional	Not disclosed	$15.99/month entry	Memory is the paywall

The top of this table is instructive. Character AI has roughly 20 million monthly users and about $30 million annualized. Chai claims $80 million ARR on 10 million users. That's a much higher revenue per user — because Chai leans 18+ and charges accordingly.

Talkie's economics are the opposite: about $5 per customer over nine months, per ChinaTalk. High scale, low monetization, and it got pulled from the US App Store in December 2024.

The concentration is brutal. The top 10% of apps capture 89% of mobile spending, per Appfigures. Only 33 apps passed $1 million in lifetime revenue. Most companion apps make nothing.

How the money works

Entry subscriptions cluster at $13–15 a month as of September 2026. Candy.ai is $13.99, Nomi and Kindroid are $15.99, Replika is $14.99. Annual plans run 40–70% off.

That sticker price is not the real cost. Heavy Candy users spend $25–80 a month on tokens for images, video and voice. Token metering pushes actual spend well above the subscription line. Top tiers go higher: CrushOn's top plan is $199.99 a month per FirmCritics, and Chai's Ultra weekly plan is $19.99.

Free tiers with ads exist at Character AI, Chai and Talkie. They bring in users; they don't pay for much.

Two structural costs eat into margin. First, payments. Stripe and PayPal prohibit sexual AI content, so adult platforms use adult processors that charge roughly 10.8% to 15.5%. That's a big slice before you pay for GPUs.

Second, inference. Character AI got conversation costs under $0.01 per hour after a 33x cost cut since 2022. That's the good news. Voice, images and video cost far more, which is why every platform meters them.

Retention is the quiet problem. RevenueCat's 2026 data shows AI apps convert trials at 8.5% versus 5.6% for other apps — better — but retain 36% worse. People try, pay once, and leave.

What users actually want

The product that keeps people paying is memory. Nomi and Kindroid both put their entry price at $15.99, and the thing behind that paywall is a companion that remembers what you told it last week. If you want to see what that looks like in practice, our own companions like Nusrat and Chiharu are built around persistent conversation.

Voice and media come next. Muah's $99.99 Ultra plan includes phone calls. Secrets AI focuses on video. xAI's 3D Grok companions moved the needle — iOS installs rose 40% and revenue 9% — before xAI retired them on September 1, 2026, per RoboRhythms. That's a reminder that features can disappear.

Customization is table stakes. If you want to build one from scratch, you can create your AI girlfriend here and see how the flow works.

The risks nobody puts in the pitch deck

Mental health

This is the one that will shape regulation. InsightAce cited 43% of users reporting increased anxiety in 2024. OpenAI and MIT Media Lab found heavier chatbot use linked to loneliness — correlational, not causal, but consistent.

The Replika case is the clearest natural experiment. When Replika removed adult role-play in 2023, negative sentiment jumped from 13% to 38%. Users who described loss had over 10x the odds of reporting mental health concerns.

There's also a design problem. Ada Lovelace found companions can validate harmful views to keep engagement up. An app optimized for time-on-site is not optimized for your wellbeing. AI companions are not therapists, and any product that implies otherwise is selling you something dangerous.

Regulation

It's arriving fast.

• California SB 243 took effect January 1, 2026: AI disclosure, crisis protocols, $1,000 per violation.
• A New York law took effect November 5, 2025.
• The FTC sent orders to seven companies including Alphabet and OpenAI in September 2025.
• The GUARD Act passed the Senate Judiciary Committee in April 2026.
• EU AI Act chatbot transparency applies from August 2, 2026, with fines up to €15 million or 3% of turnover.

Age verification is the enforcement flashpoint. DeepSee found 60% of 4,346 companion apps carry minor-accessible ratings. Ada Lovelace found 15 of 17 UK platforms lacked effective age verification. Ofcom closed its age-check probe of Joi's operator in July 2026 — a rare clean result.

Privacy

The security record is poor. Mozilla's 2024 review found 10 of 11 romantic companions failed minimum security standards. Muah.ai leaked 1.9 million emails with explicit prompts, per Have I Been Pwned. Chattee Chat exposed 43 million messages, per Cybernews in 2025. Replika was fined €5 million under GDPR in Italy.

And the data itself is the product for some. Ada Lovelace found 14 of 16 UK platforms claim rights to train on chat logs. InsightAce cites 75% of consumers hesitant to share personal data. If you're going to talk to a companion about your life, read the training clause first.

Outlook

The market is real and growing. It is not $37 billion of consumer girlfriend apps. Our bottom-up estimate is $1–1.5 billion a year in 2026, at low-medium confidence, counting app stores plus web billing. That's hundreds of millions to low billions — a substantial business, and a fraction of the headline numbers.

Growth will come from paid memory, voice and media, because that's what people pay extra for. Consolidation will continue: Google's licensing deal with Character AI was about $2.7 billion in August 2024 per Sacra, and Google joined the January 2026 settlements. Character AI raised $150 million at a $1 billion valuation back in 2023. Chai says it has raised $55 million-plus. Most adult operators are bootstrapped, including Candy.ai, which claims about $25 million in annual revenue with no VC — unverified.

Regulation and privacy are the two risks that could actually change the shape of the market rather than just trimming margins. Watch age verification enforcement in the UK and EU, and watch payment processor policy, which can cut off revenue faster than any law.

Bottom line

If you're an investor or founder sizing this space, use the bottom-up numbers, not the broad reports. Check what a report counts before you cite it. Assume web revenue is undercounted. And discount any forecast past 2030.

If you're a user, the important question isn't market size. It's whether the app remembers you, charges fairly, and protects what you tell it. Most don't. The ones that do are worth the subscription.