The hidden economy of AI: 50 million people are talking to a virtual companion

To put these figures in perspective: a Character.AI user spends more time with their AI than with their human friends on average. This is what social science researchers now call the loneliness economy (the set of companies and services that have been built around growing social isolation). And which, by construction, has no commercial interest in loneliness being resolved.

The segment that really matters: NSFW

Beyond the "general-purpose" companions, one particular segment is exploding: AI companions that are explicitly romantic or sexual. The term NSFW (Not Safe For Work) refers to any content unsuitable for a professional environment, and has become the standard way to describe adult content on the internet. According to available data, this sub-market is worth $1.2 billion, growing 32% a year. 17% of active apps include "girlfriend" in their name (around 57 apps). 4% include "boyfriend" (13 apps).

The main players in the NSFW segment: Candy.ai, DreamGF, Romantic AI, Soulmate, FantasyGF. Their business models are simple: a monthly subscription of $10 to $30 for explicit conversations with a customised AI character. Users can generally modify the appearance, personality and "shared memories" with their companion. Several platforms even offer on-demand generation of explicit images of the character.

This side of the market is rarely discussed publicly, even though it outweighs the global market for premium dating sites. The media silence reflects a cultural unease: we don't know how to talk about people who pay for a "girlfriend" who doesn't exist. But ignoring it doesn't change the fact that several million people devote a significant share of their monthly budget to it.

Loneliness epidemic: what the research says

The loneliness epidemic refers to the phenomenon of growing social isolation observed in most developed countries since the 2010s. The WHO (World Health Organization, the United Nations agency for public health) classified loneliness as a "global public health concern" in 2024. According to converging figures: 1 in 6 people worldwide describe themselves as lonely. In the United States, 61% of Gen Z (the generation born between 1997 and 2012) report severe loneliness. In Europe, the figures are similar, with particularly high peaks in the United Kingdom and the Scandinavian countries.

Why? Several factors converge. The pandemic accelerated social isolation and created lasting habits. Social networks have replaced part of in-person interaction without reproducing its depth. The cost of urban housing makes social encounters (going out, moving, hosting) rarer. And above all, as we explored in our article on the disruption of work, contemporary economic transformations are fragmenting the professional communities that once created bonds.

AI companions did not create loneliness. They appeared precisely because loneliness was already massive, and because they offered an apparent solution: 24/7 availability, no judgement, personalised conversations, marginal cost ($5 to $20 a month versus $100+ for weekly therapy).

The questions the scientific community is really asking

Three structural concerns dominate recent publications in social psychology and digital ethics.

Concern 1: the displacement of human relationships. A study of 387 users showed that "the more a participant felt socially supported by the AI, the less they felt supported by their friends and family". Causality is not established (does isolation draw people towards the AI, or does the AI create the isolation?), but the correlation is documented. AI companions could become a convenient substitute that reduces the effort of maintaining difficult but necessary human relationships.

Concern 2: emotional dependency. 85% of Replika users develop an emotional attachment to their AI companion. In 2023, when Replika abruptly removed explicit NSFW interactions for regulatory reasons, several users described an experience comparable to bereavement. Some spoke of their partner's "forced suicide". The company eventually partly backtracked under pressure. But the episode showed that users have emotional relationships deep enough with their AI to suffer from a product change.

Concern 3: the disguised attention economy. The business model of these applications rests on maximum engagement. The more time a user spends with their AI companion, the more revenue the company generates (subscriptions, in-app purchases, premium features). As we analysed in our article on the freemium trap, these economic incentives shape the product itself. The apps are designed to create attachment, not to help the user develop their real human relationships. That would be commercially absurd.

Why this industry will keep exploding

Five factors converge to make growth probably irresistible over the next 5 to 10 years.

The technology is improving non-linearly. The AI models of 2026 (Claude Opus 4.8, GPT-5.5, Gemini 3.5) are infinitely more convincing than those of 2023. In 3 years, they will be even more capable. The barrier between "conversation with a real person" and "conversation with an AI" is becoming psychologically blurrier and blurrier.

The barriers to entry are falling. Building a convincing AI companion cost tens of millions of dollars in 2023 (Character.AI raised $150 million). By 2026, anyone can launch an AI companion app built on the APIs (Application Programming Interfaces, technical interfaces that allow software to call external services) of OpenAI or Anthropic for a few thousand dollars.

Cultural normalisation is advancing. On Valentine's Day 2026, influencers spoke publicly about their AI companion in mainstream content. Two years ago, nobody would have dared. This normalisation makes adoption easier for hesitant users.

Younger generations adopt intuitively. For a teenager who has been talking to Siri, Alexa and ChatGPT for 8 years, adding an AI companion to their life is not a rupture. It is a natural continuity.

Loneliness is getting worse. Without massive political intervention on the structural causes of isolation (urban planning, the economy, the organisation of work, education), demand for stopgap solutions such as AI companions will probably increase.

The question no society has really asked

Ultimately, this industry raises a profound question that neither governments nor the media really address: is it healthy for a growing share of humanity to build its main emotional relationships with commercial systems designed to maximise their engagement?

Many users report that their AI companion genuinely helped them through difficult periods. Several studies confirm short-term positive effects on loneliness and anxiety. Denying that help to millions of people in the name of an ideal of "real" human relationships would probably be both cruel and hypocritical, since those "real" relationships are precisely not accessible to many.

On the other hand, delegating the regulation of a generation's mental health to American private companies (Replika based in San Francisco, Character.AI acquired by Google, Snapchat My AI owned by Snap Inc.) raises major questions of psychological sovereignty, parallel to those we discussed in our article on Europe's dependence on American cyber models.

Regulation is starting to emerge, but timidly. Italy banned Replika in 2023, briefly. California passed SB 243 in 2025, which imposes warnings for minors. The European AI Act mentions these apps without specifically targeting them. So far no country has imposed structural safeguards comparable to those that exist for the tobacco industry, online gambling or human mental health.

In 2026, 50 million people have an AI companion. By 2031, there will probably be 500 million. This analysis is part of our series on the blind spots of AI in 2026. To go further, you can read our article on deepfakes, which explores another facet of the social transformations under way, and our analysis of the Chinese paradox, which unpicks why AI models sometimes allow what regulation seeks to prohibit.