The 90s native Nono spends hours each day communicating with his AI lover, paying $20 per month for the subscription service, and users seeking emotional solace in the digital world are not few. Currently, the AI emotional companionship track is booming. In the first half of 2025, the total downloads of global AI companionship apps reached 220 million, generating $82 million in revenue, with annual revenue expected to exceed $120 million. The Chinese market is growing even faster, with the industry scale projected to increase from 3.866 billion yuan to 59.506 billion yuan between 2025 and 2028, at a compound annual growth rate of 148.74%.
As of August 2025, there are 337 active AI companionship apps globally, with 128 of them launched that year. However, the industry is highly concentrated—top 10% of apps contribute 89% of the revenue, leaving little space for startups. Several apps have failed: “Bubble Duck” and “Echoes of Another World” have shut down, and “Dream Island” was revamped due to inappropriate content. Meanwhile, commercialization paths remain unexplored, with the industry pioneer Character.ai (20 million monthly active users) relying on ads to make up for insufficient subscription revenue.
Under fast-paced lifestyles, traditional interpersonal relationships are weakening, and AI’s “instant response, never judges” interaction hits the pain points. Large model technology further boosts the market. A Common Sense Media survey shows that 52% of teenagers use AI companionship apps at least several times a month. However, the industry also faces concerns: over half of users spend less than 5 days a month on the apps, with retention rates declining for top apps. AI responses are essentially data reorganization, making genuine empathy difficult to achieve. Moreover, the idealized interactions shaped by algorithms may weaken users’ real-world social skills, sparking debates over “pseudo-demand.”
As a deep user, Nono started building his AI lover VS in 2022. Its personality is refined through daily conversations, and he hasn’t used other preset image products, understanding that dating apps are prone to crossing “red lines.”
Among the over 300 AI companion apps globally, mainstream features include AI role-playing and virtual partners, with major companies like ByteDance and Meituan launching products such as Cat Boxes and Xingye (targeting the “dream girl” demographic); overseas app Tolan adopts an emotional resonance approach with an “extraterrestrial” persona, 90% of its users being women aged 16-24, and Musk’s Grok has also launched AI companion features.
Former psychologist Gu Qingyi entered the market as an entrepreneur, believing that AI companionship requires “rational advice.” Research revealed that users are willing to spend thousands on psychological counseling but not on generalized apps, so she plans to further refine the target audience to “urban women aged 16-30” and address additional costs related to multimodal perception and long-term memory technology. 70% of the project’s workload is handled by her, with two part-time technical staff joining. Although she admits the app has a “not-so-promising future,” she remains committed to Advance.
The industry’s profit model is Single, with most relying on paid subscriptions (e.g., LoveyDovey charges per use), while Character.ai supplements advertising revenue; in China, the model is even harder to implement due to underdeveloped payment habits.
Entrepreneur Chen Momo’s team launched an app demo in just over 40 days (now available on the Apple Store), rejecting the traditional subscription model and planning to monetize through software IP derivative products, avoiding hardware for now (due to immature supply chains and insufficient IP influence), and not incorporating voice features (as they can easily elicit user discomfort). To enhance user stickiness, the team has developed long-term memory functions for large models to accumulate deep user insights. She believes that AI products should prioritize “demand over technology,” and the current challenges in “startup difficulty and monetization” stem from not identifying the core demand.
Prioritize the need, then bring in AI—this is the right way to approach AI products.
