Character.AI is on the verge of bankruptcy, yet the AI companionship industry has quietly started raking in massive profits.

Who could have imagined that Character.AI, once the most promising AI companion unicorn, would eventually end up filing for bankruptcy.

On October 12, Character.AI officially filed for Chapter 11 bankruptcy protection under the U.S. Bankruptcy Code, planning to sell the company's assets under court supervision. If no suitable buyer is found in the end, this once globally popular AI application may even face shutdown.

It should be noted that Character.AI is no unknown player.

It once had more than 20 million monthly active users, with users spending an average of two hours on it every day. In 2023, less than two years after its establishment, Character.AI secured $150 million in financing led by a16z, reaching a valuation of $1 billion.

At that time, it was almost one of the most successful representatives of consumer-facing AI applications.

What is even more surprising is that while Character.AI is mired in bankruptcy, a number of little-known AI role-playing applications have started making big profits quietly.

According to data from Sensor Tower, the mobile revenue of the AI companion track surged from 316 million yuan in the single month of January 2026 to 656 million yuan in August 2026, doubling in total.

Among them, Zeta, a South Korean AI role-playing application, generated 70.9 billion won in revenue in the first half of this year alone, equivalent to about 340 million yuan, and has already turned profitable.

Another AI social application named CHAI disclosed in October this year that its annualized revenue run rate has reached 120 million U.S. dollars, with a team of only a dozen to 20 people.

Products including Tipsy, PolyBuzz and Crushie are also rising rapidly on the global AI application revenue rankings.

They are all AI companion products, so why did the earliest industry leader that broke into the market go bankrupt, while latecomers start making money? Behind this may lie a major transformation in the business model of the AI companion industry.

With 20 million MAUs before filing for bankruptcy, where exactly did Character.AI go wrong?

Three years ago, Character.AI was definitely one of the most popular companies in the AI industry.

In March 2023, Character.AI secured $150 million in financing led by top Silicon Valley venture capital firm a16z, reaching a valuation of $1 billion and becoming one of the earliest AI application companies to enter the unicorn club.

Less than a week after the launch of Character.AI's mobile application, its downloads exceeded 1.7 million times.

In September of the same year, according to data from Similarweb, Character.AI reached 4.2 million monthly active users on U.S. mobile devices, close to the roughly 6 million MAUs of ChatGPT's mobile version in the same period.

By 2024, Character.AI had become the most popular AI companion application in the world.

According to data from Sensor Tower, in August of that year, the global monthly active users of Character.AI hit a record high of 22 million, ranking first in the AI companion track.

What is even more remarkable is user stickiness. According to statistics from Sensor Tower, active users of Character.AI spend more than 90 minutes on the platform on average every day, initiating about 25 conversations on average.

What does this mean? At that time, even TikTok, which is famous for its high user stickiness, only had an average daily usage time of a little more than 70 minutes.

Why are AI companions so attractive?

Noam Shazeer, founder of Character.AI, once put forward a very interesting view: The global entertainment industry is as large as 2 trillion U.S. dollars every year, and one core essence of entertainment consumption is that humans build emotional connections with fictional characters.

They believe that the interaction process itself can also become a kind of experience with commercial value.

Powered by large language models, virtual characters can not only have free conversations with users, but also continuously adjust their interaction methods based on users' preferences and experiences. This means that entertainment content that users could only passively watch in the past has begun to turn into two-way interaction that users can continuously participate in.

Although Character.AI proved that users are willing to spend a lot of time building emotional connections with AI, it never managed to answer a more important question: how on earth can this time be converted into revenue?

Looking back, the failure of Character.AI actually exposed two contradictions in the early business model of the AI companion industry.

On the one hand, a large number of the platform's most popular characters come from animations, games and film and television works. Character.AI attracted tens of millions of users relying on these IPs, but it does not own the copyright of many of these characters. This means that although it controls the traffic entry, it does not fully hold the pricing power of the content.

On the other hand, AI companions are different from traditional internet products. Every extra sentence a user chats means real model inference cost behind it. However, in the early stage, Character.AI mainly monetized through a $9.99 monthly membership subscription, and a large number of free users could chat unlimitedly.

The result is that the more time users spend on the platform, the higher the cost grows, but the revenue does not increase synchronously.

According to bankruptcy filings, as of August 2026, the monthly revenue of Character.AI only reached about 5.3 million U.S. dollars, including 2.8 million U.S. dollars in subscription revenue and 2.5 million U.S. dollars in advertising revenue.

In August 2026, the company's monthly operating expenses were about 13.3 million U.S. dollars, and the monthly loss still reached 8 million to 9 million U.S. dollars. Among them, the computing power expenditure alone reached 5.61 million U.S. dollars, accounting for about 45% of operating expenses.

The business model of emotional companion products like Character.AI is a bit like a gym: it can only tolerate light usage from users, and heavy usage will lead to losses. This turns into a dead end: heavy users consume a lot of resources, while light users have poor retention.

At the same time, Character.AI faced multiple lawsuits and regulatory investigations due to safety issues caused by minors using its AI chatbots.

These problems not only brought high legal costs, but also further increased the difficulty for the company to raise funds and continue operations.

A series of troubles eventually pushed Character.AI to bankruptcy.

The AI companion sector is not cooling down, instead it is quietly making huge profits

The bankruptcy of Character.AI did not make the AI companion track fall into silence. On the contrary, a number of new AI role-playing products are making big profits quietly.

According to statistics from Appfigures, a mobile app data institution, in the first half of 2026, 214 global romance and adult-oriented AI companion applications generated a total of 162.8 million U.S. dollars in mobile user consumption.

Among them, South Korea's Zeta ranks first with 33 million U.S. dollars, Tipsy Chat follows closely with 15.2 million U.S. dollars, and Crushie AI also reached 8.8 million U.S. dollars.

It is worth noting that some products have already achieved profitability.

For example, Scatter Lab, the South Korean company behind Zeta, recorded 70.9 billion won in revenue in the first half of this year, 2.7 times the 26.7 billion won revenue for the whole of last year, and posted an operating profit of 5.2 billion won at the same time.

This means that the AI companion sector has begun to prove itself as a truly profitable business.

Why did these latecomers manage to make money that Character.AI failed to earn? One important reason is that they no longer treat AI companions simply as chat tools, but operate them more and more like a content business.

Traditional AI companion products have long had two unavoidable shortcomings.

First, the emotional experience is not realistic enough, and it is difficult to sustain.

AI can hardly fully understand and simulate complex human emotions, and problems such as memory loss and inconsistent character settings often occur during conversations. After chatting for a while, users will find that the character in front of them no longer seems to be the person they knew at first.

The other problem is the excessively high interaction threshold.

In the past, under the logic of social networking, most AI companion products stayed in the chat box. Users needed to constantly initiate topics on their own, build scenes and push the plot in their minds through back-and-forth chats, and then gradually invest their emotions.

To put it bluntly, users not only need to participate in the story, but also have to take on a considerable part of the content creation work themselves.

But the problem is that only a small number of users are willing to spend a lot of time creating stories. For most people, they are obviously more accustomed to consuming stories than writing stories on their own.

And the biggest change of this new batch of products now is to let AI take on more content production work, turning users from story creators into participants and consumers of stories.

The most representative one is Tipsy.

Different from Character.AI which mainly provided character chats in the early stage, Tipsy puts more emphasis on interactive plots. Users can directly choose their favorite characters, enter a virtual world with pre-set character relationships and story backgrounds, and push the plot forward in the process of interacting with AI.

For example, users can play the protagonist in a romance story, change the character relationship through different choices, and even affect the direction of the whole story.

At the same time, the platform can also generate images and videos based on the plot, so that users no longer have to rely solely on text to imagine characters and scenes. The whole experience is getting closer and closer to an interactive game that can generate plots in real time.

South Korea's Zeta is also developing along a similar direction.

In the past, Zeta mainly relied on one-on-one chats between users and virtual characters. But starting from this year, it has gradually shifted the product focus to interactive stories that include multiple characters, complex character relationships and a complete worldview.

In March this year, Zeta even adjusted the content unit on the platform from "character" to "plot", further emphasizing the importance of the story itself.

According to the official explanation, the content on the platform now is no longer limited to a certain virtual character, but has gradually evolved into a story that includes protagonists, supporting roles, character relationships and a complete worldview.

To further lower the threshold for users to participate in the plot, Zeta launched the AI plot option feature in April this year.

For example, after users enter a campus romance or fantasy adventure story, they no longer need to constantly find topics and write dialogues on their own. AI can automatically generate different options based on the current plot, and users only need to click to decide the direction of the subsequent story.

Behind these several adjustments is actually the same idea: let AI take on more story creation and plot advancement work, reducing the energy users need to actively invest. Such changes are obviously closer to the game and content consumption experience that the public is familiar with.

As the product form changes, the monetization methods of these companies are also starting to move closer to traditional games and content platforms.

Tipsy is increasingly acting like a mobile game company: attracting users through characters and plots, and then increasing the consumption amount of a single user via subscriptions, virtual currencies and value-added interactions.

According to a study by Airbridge in October this year, Tipsy offers three tiers of monthly subscription services at $4.9, $14.9 and $44.9, and also sells virtual gems, with the highest-tier gem pack priced at $199.

These gems can be used for features such as character chat, image and video generation. Using more advanced models, generating more content, or engaging in more complex long conversations may lead to additional consumption.

This mechanism is actually very similar to monthly cards, diamonds and paid props in mobile games.

Users can enter the story for free, but if they want to get a better interactive experience, richer audio-visual content, or continue to dive into a certain plot, they have reasons to make further payments.

Compared with traditional AI chat products that mainly charge through fixed subscriptions, this model has another important advantage: the amount users spend can be linked to the depth of experience.

Summary

Looking back, the biggest difference between Character.AI and these latecomers is that the latter have a clearer understanding of exactly why users pay.

Character.AI tried to create a borderless virtual emotional relationship, but the more addicted users are, the more money the platform may lose. In contrast, Zeta and Tipsy choose to return to the logic of games and content consumption, turning emotional companionship that was originally difficult to price into an entertainment experience that can generate continuous payments.

This also means that AI companions are moving from a grand story that tries to subvert human social relationships back to a content business with a clearer business model.

Although this story is no longer as grand as initially imagined, for the AI companion industry, finding a truly profitable path may be far more important than telling a bigger story.

This article is from the WeChat official account "Silicon-based Observation Pro", author: Silicon-based Jun, published by 36Kr with authorization.