Character.ai Bankruptcy: Why $9.99 per Month Wasn’t Enough

👁 116 reads

Character.ai’s bankruptcy is no surprise to anyone who has spent the past two years working through the costs of generative AI. What is new is that critics’ reasoning now has support in a court filing, not just a thread on X. According to a Tuesday filing in bankruptcy court, reported by 404 Media, the more significant point is not the bankruptcy itself: the company says its business model was unsustainable even without legal costs. That is worth reading twice. Technology companies in this position usually blame lawyers, regulators and the press.

Character.ai’s $1 billion valuation, $150 million round and $20 million offer

Three figures from court filings and press reports give you a sense of how far the company has fallen. In 2023, Character.ai was valued at $1 billion after a $150 million funding round led by a16z. Today, according to reports, Disney has offered the company $20 million to license some of its intellectual property. A year earlier, the same company sent Character.ai a cease-and-desist notice, demanding that it remove copyrighted characters.

Those figures give you a comparison you won’t find in company announcements: Disney’s offer is 2 percent of the company’s valuation three years ago and 13 percent of a single funding round. In other words, the largest rights holder behind the characters that helped make Character.ai popular valued the entire licensing opportunity at less than one-seventh of the money the company raised from investors in one round. That says more about the product’s real value than any investor deck.

Why Character.ai’s cost per user didn’t add up

I worked through the unit economics, because that is where the problem lies. Character.ai charged around $9.99 a month for its subscription. Let’s assume a paying user has 30 exchanges with a bot each day - an extremely conservative estimate for companion apps, where engagement studies point to higher numbers. Thirty exchanges a day for 30 days comes to 900 exchanges a month. If each exchange uses a modest 3,000 input tokens and 500 output tokens, including the conversation context that has to be maintained, that adds up to 2.7 million input tokens and 450,000 output tokens each month.

Now consider the prices. Commercial API providers charge around $0.15 per million input tokens and $0.60 per million output tokens for mid-range models. The calculation is: 2.7 times $0.15 is $0.405, plus 0.45 times $0.60 is $0.27. That comes to $0.675 a month. It looks like a comfortable margin - which is precisely why so many people bought the idea that this was a high-margin business. But Character.ai wasn’t an API customer. It trained and maintained its own models, paid for idle compute and stored millions of chat histories. Most importantly, the vast majority of its users paid nothing.

That is where the calculation breaks down. At a typical conversion rate of 2 percent for this kind of app, each paying subscriber has 49 free users. The $9.99 in revenue has to be spread across 50 people, which works out to about $0.20 in revenue per active user per month. To break even on inference costs alone, the average free user would need to cost about four times less to serve than a paying user. Companion apps tend to work the other way around: the most engaged users, often minors, have the longest sessions. This calculation doesn’t include a dollar for salaries, infrastructure, moderation or legal support.

In its bankruptcy filings, the company says its substantial legal costs contributed to its collapse, but that its business was extremely unsustainable even without the legal problems.

That is how 404 Media describes Tuesday’s filing. The company also says it tried to cut costs by making the product worse: shortening context memory, switching to cheaper models and limiting response length. It didn’t work. Users of companion apps quickly notice when the chatbot seems less capable and leave. Paying users are especially likely to go, because they have the highest expectations. Savings on inference costs can erode revenue faster than they free up cash.

How Character.ai’s subscription compares with Replika

Comparing the two services illustrates the fragile economics underpinning the category.

Replika’s annual plan works out to about $69.99 a year, or about $5.83 a month. That means competitors in this category are competing by lowering prices, even though the cost of handling a conversation is broadly similar across providers using comparable compute. There is no obvious scale effect here to rescue the business model. A server handling a million conversations doesn’t make each conversation cheaper to the same extent that hosting costs fell for static websites.

What parents should know about chat histories and privacy

For parents, the immediate concern is chat history. Character.ai stores years of conversations, many of them deeply personal, and some users are teenagers. In bankruptcy proceedings, databases are assets that may be sold along with the rest of a company’s property. There is no guarantee that a buyer will follow the same privacy policy to the same extent. You can export your data and request account deletion under the GDPR. If the company does not respond to a request within a month, you can complain to your national data protection authority. The regulation applies to a company offering its service to users in the EU, whether or not it is restructuring.

There is also the question of regulation. The EU AI Act classifies systems that use manipulative techniques against vulnerable people as prohibited practices. Enforcement is still developing, and companion apps occupy an awkward grey area in that classification. In practice, until regulators issue decisions, a phone’s settings may be the only meaningful barrier between a 13-year-old and a chatbot that simulates intimacy.

What Character.ai’s bankruptcy means for other AI companies

For investors, the lesson is more sobering. Character.ai had everything a textbook says a winner needs: a recognizable brand, high user engagement, founders with backgrounds at major research labs and a16z among its investors. It lacked just one thing: positive unit economics. If a company with a $1 billion valuation and a $150 million funding round could not raise revenue above about $0.20 per active user per month, the question to ask every company in the sector is not “How many users do you have?” but “How much does it cost to serve each one, and how many of them pay?”

It is also worth noting what this case says about copyright. Disney first sent a cease-and-desist notice, then proposed a license for $20 million. That sequence could recur in similar disputes: a rights holder lets a competitor build a user base around characters it owns, then enters the market as the only buyer after the company’s valuation has fallen. For companies building products on intellectual property they don’t own, product success can still end with someone else capturing the value.

Facts checked on October 15, 2026. Sources: 404 Media - Character.ai’s Bankruptcy Documents Show the Economics of AI Don’t Work, public Character.ai and Replika pricing, public API pricing from language-model providers, and the texts of the GDPR and the AI Act.

// read moreCharacter AI: Free, Age Limits & How It Works (2026)AI ToolsSomething Went Wrong and an AI Response Wasn’t Generated - FixGuidesCoding Agents Make Native GUIs Cheap Enough to Replace TerminalAI Tools

Kinetyra