Behind Character.AI's Bankruptcy: Why AI Companionship Still Struggles to Make Money Despite Revenue Growth

This star company, which reached a valuation of $1 billion in just over a year of existence and has tens of millions of monthly active users, was once regarded as one of the most representative players in the AI companionship sector. With a product that lets users chat with virtual characters, it attracted a large number of users and capital, yet ultimately ended up in bankruptcy reorganization.

On October 12, Character.AI's parent company filed for bankruptcy protection. Sustained computing power investment, operating costs, and litigation and regulatory pressure related to the protection of minors plunged this star company into operational difficulties. Currently, Disney has emerged as the first bidder with $20 million, bidding for a non-exclusive perpetual license to part of its intellectual property. A star company once valued at $1 billion now has no choice but to dispose of some of its assets through bankruptcy proceedings.

But the fall of Character.AI does not mean the AI companionship market has lost its growth momentum. Another AI character chat platform, Chai, disclosed that as of early October, its daily active users exceeded 1 million, and its annualized revenue reached $120 million. According to Sensor Tower data, in the first quarter of 2026, global revenue from AI companionship mobile apps reached about $150 million, more than 12 times that of the same period in 2023.

The domestic market also has gathered quite a few players. From MiniMax's Xingye and Talkie, to ByteDance's Maoxiang, and to a batch of startup products focused on niche needs, companies from different backgrounds are all exploring business opportunities in AI companionship.

01. AI companionship: some go bankrupt, some grow

The overseas market was the first to produce a batch of star products, and also the first to begin facing the test of commercialization.

Character.AI once built a huge user base through virtual character chats, but its user scale did not translate into revenue sufficient to support operations.

According to Reuters, the company spent about $13.8 million per month on computing power in 2024, which dropped to about $10 million in 2025, and totaled about $56.1 million from January to August 2026. On the revenue side, monthly subscription revenue grew from about $600,000 in January 2024 to about $2.8 million in August 2026; advertising revenue grew from about $500,000 in July 2025 to $2.5 million in August. Total revenue in August was about $5.3 million, still insufficient to cover its operating expenses.

It is worth noting that Character.AI adopted a "free + paid" model, with the paid version divided into a light membership at $4.99 per month or $2.49 per week, and a premium membership at $9.99 per month or $94.99 per year. The lower-priced model was only launched in September 2026 to ease the financial pressure of continued losses. But by the time it filed for bankruptcy, it was still losing $8 million to $9 million per month.

Chai, another platform also focused on AI character interaction, is in a different situation. As of early October 2026, its daily active users exceeded 1 million, and its annualized revenue reached $120 million. Chai claims that the company has achieved a positive unit economic model and is profitable after excluding sales and marketing expenses. However, these figures are unaudited and do not mean the company's overall net profit has turned positive.

Chai's charges are divided into consumer subscriptions, in-app purchases and advertising. According to data cited on its official website, during the period from September 10 to October 9, 2026, its app store net revenue was $5.08 million (mainly covering subscription and in-app purchase revenue). According to company data, about 60% of its revenue comes from subscriptions and in-app purchases, and 40% from advertising.

In addition, earlier entrants to the market such as Replika are still operating. Different products differ in user scale, revenue level and operating results.

The first category is large model manufacturers, represented by MiniMax. It has successively launched AI character interaction products such as Xingye (domestic) and Talkie (overseas), applying underlying model capabilities directly to the consumer market, mainly monetizing through subscriptions and advertising.

AI companionship applications were once an important support for MiniMax's early commercialization. According to previously disclosed company data, in the first nine months of 2025, Talkie and Xingye together generated about $18.75 million in revenue, contributing about 35.1% of revenue, surpassing Hailuo AI's 32.6%. Among them, Talkie was also one of MiniMax's important products for opening up overseas markets.

But this business does not mainly make money from membership subscriptions; it relies more on online marketing services such as advertising. In the first nine months of 2025, the two products together generated $18.75 million in revenue, of which online marketing services revenue was $11.188 million, accounting for about 59.7%; subscription revenue was $6.604 million, accounting for about 35.2%; and in-app recharge revenue was $958,000, accounting for about 5.1%.

Even though advertising accounts for the bulk, that does not mean the related business has a high gross margin. Looking at MiniMax's broader AI-native product business, including Talkie, Xingye, Hailuo AI and other products, the gross margin was -8.1% in 2024; it only turned positive in the first nine months of 2025, at just 4.7%. But note that this is not the gross margin of Talkie and Xingye alone.

In terms of user scale, as of August 2026, Talkie had about 31.07 million global monthly active users, and Xingye about 3.38 million. However, some products have also seen a marked decline in new downloads. Diandian Data previously showed that from January to May 2025, Xingye's monthly downloads on the Apple side fell from 4.86 million to 930,000, a drop of about 81%, but daily active users in January and May were about 960,000 and 970,000 respectively, remaining basically stable.

More noteworthy is that MiniMax's revenue focus is shifting. In the first half of 2026, the company's AI-native product revenue was $42.64 million, and its share of total revenue fell from 69.7% in the same period last year to 36.6%; the share of open platform and enterprise service revenue rose to 63.4%. For this large-model company, which first opened up the market with AI companionship, the growth driver has mainly come from the enterprise side.

The second category consists of those incubated by internet and content platforms, such as ByteDance's Maoxiang and Yuewen's incubated Zhumengdao (now operating independently). Such companies have advantages in traffic entry points, product operations and user reach, and can integrate AI character interaction into the existing consumer internet ecosystem.

Sensor Tower data shows that in the first quarter of 2026, Maoxiang ranked first on China's iOS AI application revenue list, and its cumulative iOS in-app purchase revenue has exceeded $22 million, which does not include Android channels and advertising revenue. In August 2026, calculated by iOS revenue generated per monthly active user, Maoxiang was about 5 times that of Jimeng AI, 4 times that of AI Mate, and also about 4 times that of Xingye and Zhumengdao.

Maoxiang's way of making money is also different from the traditional membership system. In addition to subscriptions and advertising, there is also recharge of virtual currency. Maoxiang's continuous monthly membership is 25 yuan, and users can get benefits such as longer voice call time and more advanced conversation modes; in addition, users can also buy a virtual currency called "cat food," with 100 cat food priced at 1 yuan, used to unlock some plot chapters, character updates and other content, and cat food can also be obtained by watching advertisements. It provides AI companionship with a revenue source beyond subscriptions, but which method contributes more revenue specifically has not yet been officially disclosed.

The third category is independent startups. Such players usually do not have the traffic base of big companies, nor do they necessarily have self-developed large models, so they rely more on product positioning, user needs and operational efficiency to find market space. For example, Xinying Suixing, which received tens of millions of dollars in A+ round financing, launched Doudou AI, which focuses on real-time companionship in games; Solulu, which has original IP and focuses on AI chat companionship; and Tianzhiling AI, mentioned below.

Such players are usually smaller in scale and more focused in product, and compared with the first two categories, the common problem faced by independent startups is how to attract users under limited resource conditions, establish a payment model, and control the continuously incurred model invocation costs.

On the AI companionship track, having technology, traffic or content resources does not mean one is certain to succeed. Large-model vendors have technological advantages, internet platforms have user bases, and content companies hold IP resources, but these advantages will not automatically translate into commercial returns, and may even become cost burdens.

02. Revenue is rising, so why is making money still difficult?

From the industry overall perspective, AI companionship is still heating up. According to data from the Zhongyan Puhua Industry Research Institute, China's AI emotional companionship segment alone reached 3.866 billion yuan in 2025, and is expected to exceed 59.5 billion yuan by 2028, with a compound annual growth rate as high as 148.74%. At the same time, in the first half of 2025, global AI companionship apps generated $82 million in revenue, with an obvious year-on-year growth trend.

But more than one entrepreneur in AI companionship applications told "AIX Finance" that the more realistic situation is that profitability remains a practical problem faced by many practitioners.

The problem lies at both ends: revenue and cost.

First, on the revenue side, users' reasons for paying for standalone applications are weakening.

Previously, capabilities such as long-term memory, role-playing and voice interaction were important selling points for AI companionship products to attract users. But as large-model capabilities continue to upgrade, these functions are no longer unique to dedicated AI companionship applications. For basic needs such as daily chatting, venting troubles and seeking emotional comfort, users can completely open general-purpose AI assistants such as Doubao and ChatGPT.

Image source / Unsplash

An AI product manager told "AIX Finance" that an unspoken industry consensus is that some AI companionship products in the past attracted users through more intimate character interaction, ambiguous relationships and even content with adult implications, and such content may be the main reason some users are willing to pay.

But in China, regulation is getting tighter, and relying on breaking through content boundaries to attract users, even if it can generate revenue in the short term, is unlikely to become a stable and sustainable business model.

This means it is increasingly difficult for standalone applications to charge users simply by "being able to chat."

The internet industry is accustomed to measuring product value with metrics such as monthly active users, retention rate and usage duration. The more users and the longer they stay, usually the greater the space for advertising monetization and the easier it is to attract capital attention. Traditional internet products mostly serve one user, and marginal costs are almost negligible.

Zhang Yi believes that directly transplanting the growth logic of the traditional internet into the AI companion industry may be a risk entrepreneurs face. For example, some entrepreneurs hope to use operational tactics to increase user activity, yet overlook that AI conversations themselves require ongoing costs. Some heavy users may send 20,000 to 30,000 messages a month; if a product cannot control the cost of a single conversation, such users may even cause the company to lose money.

Character.AI's predicament is a case in point.

According to public materials, Character.AI has about 12 million active users. In August 2026, the platform generated about $0.44 in revenue per active user per month, of which about $0.21 came from advertising, while the computing cost per user was about $0.58. The more active the users, the higher the computing bill. Add to that personnel expenses, litigation costs and regulatory pressure, and the huge user scale did not narrow the losses.

Chai also experienced similar cost challenges. Its founder, William Beauchamp, once mentioned in a media interview that in some markets, every $1 of computing power consumed by users generated only about $0.1 in revenue.

An investor focused on AI applications believes that what AI companion companies need to figure out is not only customer acquisition cost and paid conversion rate, but also the ongoing service cost generated by each user during use. They should look at the revenue contributed by users over a certain period, how much remains after deducting direct service costs such as model inference, and whether that revenue can further cover customer acquisition and daily operating expenses.

In his view, entrepreneurs also need to first think clearly about what kind of company they want to build: a small company that can be consistently profitable, or a company worthy of continued venture capital investment with the potential for scaled growth. The former only needs to find enough stable paying users and keep operating costs within revenue; the latter must further prove that user scale can expand, revenue growth is replicable, and growth will not continue to depend on heavy subsidies and computing power investment.

Of course, this does not mean user scale has completely lost its value.

For AI companion platforms that adopt an advertising model, user scale and usage duration may still bring advertising revenue. Products such as Chai are also exploring monetization methods that combine subscriptions with advertising. As long as the revenue brought by a single user can cover the cost he consumes, large-scale users may still become an advantage.

By contrast, some players have begun to make an issue of pricing. In early 2026, Chai founder William Beauchamp publicly explained that due to differences in advertising monetization capabilities across regions, Chai retains free service in markets such as the United States, Europe and Japan where advertising revenue is sufficient to cover computing costs, while temporarily setting a payment threshold in some markets such as India and Pakistan. He said the price of that package would be kept at the "absolute minimum" that only covers electricity costs, but ads would still be shown to users.

There is also the "cat food recharge" of Maoxiang mentioned earlier. Compared with relying solely on membership subscriptions, this model is closer to the payment logic of games and online literature: users not only pay for AI conversation capabilities, but can also make small purchases for characters, plots and interactive content they are interested in. By binding payment to specific content, the platform both lowers the threshold for a single user purchase and has the opportunity to gain more revenue as users' interaction needs increase.

03. The next opportunity: niche markets, products and hardware?

First, it may be those niche markets that look obscure but have real paying demand.

Tianzhiling AI mainly uses AI technology to help users converse with deceased relatives in the form of an "AI relative." Zhang Yi told "AIX Finance" that when the product first launched, the payment page was not well designed and was even hard to find, but users still proactively contacted the team to ask how to pay. In his view, the main reason is that the product hit a real and strong emotional need among users.

He disclosed to "AIX Finance" that Tianzhiling AI currently has about 111,000 cumulative registered users and about 2,000 to 3,000 paying members. The common membership price is 99 yuan per year. Although it started relatively late and the product is relatively new, it has already achieved profitability.

The aforementioned investor believes that to judge whether a niche market is worth entering, at least three questions need to be considered: whether user demand is strong enough, whether existing products have fully satisfied that demand, and whether users are willing to continue paying for a better experience. More importantly, whether the paying demand can support the company's long-term operation.

Of course, niche does not equal high willingness to pay. Although many segmented needs exist, the user base is limited, usage frequency is low, or users themselves lack the motivation to pay continuously.

Second, rather than simply competing on model capabilities, whether a product can provide services that meet user expectations may be more important.

Image source / Unsplash

Users can certainly directly ask general AI assistants such as Doubao and ChatGPT to play a certain role, but for a product specifically providing related services, beyond being able to generate answers, it also matters whether the character identity is stable, whether responses fit a specific relationship, and whether the conversation suddenly breaks immersion. Zhang Yi said that to enable AI to continuously provide responses that meet user expectations in specific scenarios, it is necessary to repeatedly test and adjust dialogue strategies and optimize prompts, while reducing model invocation costs as much as possible while ensuring the experience. "These are all things that look simple but are difficult to do."

For example, after a virtual romance companion application attracts a large number of users, some of them hope to have offline companion robots in addition to online chatting. For companies, software may also become a channel for understanding user needs and finding hardware consumers.

The aforementioned AI product manager also told "AIX Finance" that combining software and hardware may be the way to further expand the boundaries of AI companionship. In his view, AI companionship in pure software form mainly relies on text, voice and other methods to interact with users, while hardware such as AI earphones, glasses and companion robots has the opportunity to allow AI to perceive more of the real environment users are in and understand user states through touch, movement and other means.

For example, the AI companion toy Fuzozo launched by Luobo Intelligence is priced at 399 yuan for the hardware, while also providing additional paid services through the App, with monthly fees between 19.9 and 29.9 yuan. It is reported that its sales exceeded 120 million yuan in just over a year since launch. Similarly, there is the desktop AI pet robot ropet Roupaipai launched by Mengyou Intelligence, priced at about 2,000 to 3,000 yuan, with no additional subscription in the App but with a points system and nurturing tasks. According to public information, the company has raised more than $10 million, and from September 2025 to April this year it shipped nearly 20,000 units cumulatively.

However, there is still a considerable distance between software users and hardware consumers. Hardware also involves R&D, production, inventory, after-sales and other issues, and its operational difficulty is completely different from asset-light software services.

This article comes from the WeChat public account "AIX Finance," author: Li Mengran, editor: Wei Jia, published by 36Kr with authorization.