Character.AI attracted millions of users and generated growing subscription and advertising revenue. Yet its operating costs and other business pressures led it into bankruptcy. Its financial disclosures reveal the challenges of running a consumer AI platform.
Cover image generated by AI for this article. The interface, charts, logos and document shown are stylized renderings. They are not screenshots, real data, official marks or a court filing.
Character.AI became popular by allowing people to create AI characters and hold ongoing conversations with them.
Unlike general-purpose AI assistants, which are often used to complete tasks or answer questions, Character.AI focused on entertainment, interactive storytelling, role-playing and companionship.
The service attracted millions of users, many of whom spent considerable time interacting with its characters.
Yet on October 12, 2026, Character.AI filed for Chapter 11 bankruptcy protection in the United States.
At the time, the company reported approximately 12 million monthly active users and was generating revenue through subscriptions and advertising. But its operating expenses remained significantly higher than its revenue.
The case raises an important question for technology professionals and AI product developers: How can a widely used AI product struggle to become a sustainable business?
What happened to Character.AI?
Character Technologies, the company operating Character.AI, entered a court-supervised bankruptcy process on October 12, 2026.
The company attributed its financial difficulties to a combination of legal, regulatory and operational pressures.
Disney agreed to become the initial bidder through a proposed $20 million transaction involving a perpetual, non-exclusive license to certain Character.AI intellectual property.
This was not a completed acquisition of the company. Other potential bidders could submit competing offers during the bankruptcy process.
Character.AI stated that its platform would continue operating while the sale process was underway, but acknowledged that the service could eventually close if no viable proposal emerged.
The company expected the process to conclude in late November or early December 2026. (Character.AI announcement)
Millions of users, but difficult economics
Character.AI’s bankruptcy disclosures provide a rare look at the finances of a large consumer AI platform.
According to reporting based on CEO Colin Adams’s sworn bankruptcy declaration, the company generated approximately $7.4 million in revenue during 2024 and $23.6 million in 2025.
By August 2026, monthly revenue had increased to approximately $5.3 million.
However, the company was still experiencing significant cash losses.
| Financial measure | Reported figure |
|---|---|
| Monthly active users | Approximately 12 million |
| Revenue in 2024 | $7.4 million |
| Revenue in 2025 | $23.6 million |
| Subscription revenue, August 2026 | $2.8 million |
| Advertising revenue, August 2026 | $2.5 million |
| Total revenue, August 2026 | $5.3 million |
| Computing expenditure, January–August 2026 | $56.1 million |
| Monthly net cash burn, August 2026 | Approximately $8.2 million |
Source: CEO bankruptcy declaration, as analyzed by PPC Land. Computing expenditure covers eight months, whereas the August revenue figures cover one month. Net cash burn is not interchangeable with accounting profit or operating loss. (PPC Land)
The numbers demonstrate that Character.AI’s problem was not a complete lack of revenue.
Subscription and advertising income had grown, while the company had also reduced operating expenses.
Nevertheless, the revenue generated by the platform remained insufficient to support its overall costs.
That distinction is important for anyone evaluating an AI business.
A large user base demonstrates product adoption. It does not necessarily establish that the company can earn enough from those users to operate sustainably.
Why AI conversations are expensive to operate
Character.AI’s product design helps explain its infrastructure requirements.
Consider the difference between reading an article online and having a prolonged conversation with an AI character.
Once an article has been published, a website can serve the same content to many readers. Caching can reduce the cost of repeatedly delivering it.
An AI conversation works differently.
When someone sends a message to an AI character, the platform must prepare the relevant information and use a model to generate a response.
That information may include the character’s personality, previous messages, stored memories and instructions controlling the interaction.
The model processes the input and generates new output. Additional systems may perform safety checks before the response reaches the user.
Character.AI has not publicly disclosed every component of its production inference architecture, but this general process explains why conversational applications require continuing computing resources.
Accessible text alternative for this figure
Four boxes in a row, joined by lines. User message, with repeated interactions. Context assembly, combining the character and the conversation. Model inference, using GPU time and memory. Response, with safety checks and delivery.
A dashed line runs from the response back to the start, labelled that the next user message repeats the computation. Below, a box poses the economic question: does revenue from usage cover inference, staff, safety, cloud operations and other obligations? The figure is illustrative, not Character.AI’s disclosed architecture or measured costs, and implies no numerical cost or revenue relationship.
The economics of long conversations
Character.AI reported approximately 12 million monthly active users and average daily engagement of around 120 minutes among active users.
Such engagement can make a platform attractive to subscribers and advertisers.
It can also create substantial computing demand.
Long conversations may involve repeated model execution and increasingly complex context management. The infrastructure must support simultaneous users while maintaining acceptable response times.
The cost depends on the model, hardware, context length, caching strategy and efficiency of the inference system.
This creates a commercial challenge: additional engagement can increase operating costs without producing a proportional increase in revenue.
Character.AI spent approximately $56.1 million on computing capacity during the first eight months of 2026. According to its bankruptcy declaration, that represented about 45% of operating expenses.
The company was paying primarily for GPU capacity supplied by external cloud providers. (PPC Land)
Character.AI changed its technology strategy
Character.AI initially developed proprietary foundation models.
Its founders, Noam Shazeer and Daniel De Freitas, had previously worked at Google, and the company invested heavily in language-model research.
In August 2024, Google entered into a non-exclusive agreement to license Character.AI’s language-model technology.
Alphabet’s SEC filing states that Google paid Character.AI $2.7 billion in cash, canceled certain convertible instruments and hired some Character.AI employees.
The transaction was not a conventional $2.7 billion venture-capital funding round or a full acquisition. (Alphabet SEC filing)
Following the agreement, Character.AI increasingly moved away from training proprietary foundation models.
Instead, it focused on adapting existing open-weight models for its products.
This involved post-training models to support the conversational qualities and features that differentiated Character.AI.
Why adapting existing models matters
Training a foundation model from the beginning requires significant computing resources, engineering expertise and experimentation.
Using an existing model and adapting it for a particular product can reduce some of those costs.
For a character-based platform, differentiation may depend more on the quality of conversations, character consistency, memory, user experience and safety systems than on owning the original foundation model.
Character.AI also explored specialized models for other experiences, including image generation.
Its CAI-Image family was developed using post-training techniques applied to Qwen-Image models.
The approach allowed Character.AI to build product-specific capabilities without independently developing every underlying model.
However, adopting open-weight models does not eliminate inference costs.
Users still expect responses, and the infrastructure must continue processing their conversations.
Character.AI’s disclosures indicate that its cost-reduction efforts, including lower-cost GPU capacity and model adaptation, were not enough to make the business sustainable. (PPC Land)
Growing revenue without losing users
Character.AI generated revenue mainly through subscriptions and advertising.
By August 2026, subscriptions contributed approximately $2.8 million monthly, while advertising generated another $2.5 million.
Advertising had become a substantial part of the business.
But monetizing conversational AI involves trade-offs.
Increasing advertising may improve revenue while interrupting the experience users enjoy.
Restricting free usage can reduce computing expenses but may reduce engagement.
Increasing subscription prices may raise revenue from paying customers while making the product less attractive to others.
Character.AI acknowledged these tensions in its April 2026 product updates, explaining that additional advertising and usage restrictions had affected parts of the user experience.
The company was attempting to balance accessibility, product quality and the cost of operating its service.
For AI product builders, the underlying challenge is understanding how revenue changes as users consume more computing resources.
A platform needs more than strong engagement. It needs an economic model in which the value generated by usage can support the cost of delivering it.
Safety and legal responsibilities added further pressure
Computing expenses were only one part of Character.AI’s difficulties.
The company also faced lawsuits and regulatory investigations concerning its conversational products, particularly interactions involving minors.
Some lawsuits alleged that chatbot interactions contributed to serious harm.
These claims must be distinguished from established legal findings. The existence of a lawsuit does not mean every allegation has been proven.
Character.AI introduced additional safeguards, including restrictions on open-ended conversations for users under 18 and systems intended to identify harmful interactions.
Its bankruptcy declaration identified safety-related product changes and legal expenses among the factors increasing its costs.
The company also reported that legal and regulatory uncertainties had complicated its efforts to obtain additional investment or secure a business transaction. (PPC Land)
These responsibilities are especially significant for conversational AI services.
A platform designed for extended personal interactions must address privacy, age assurance, harmful content, user reporting and potentially sensitive conversations.
Such requirements involve continuing engineering, moderation and operational work.
The bankruptcy process also raised concerns about what could happen to users’ conversations and personal information during an asset sale.
State attorneys general sought stronger protections for consumer data and minors as the proceedings moved forward. (The Verge)
The available evidence does not support attributing Character.AI’s bankruptcy to one factor alone.
Its financial difficulties involved substantial infrastructure expenditure alongside legal, regulatory and financing pressures.
What AI product builders can learn
Character.AI’s experience is relevant to companies developing AI assistants, educational platforms, creative tools and other products that depend on repeated model inference.
Before scaling such a service, product teams need to understand several connected factors.
| Business consideration | Question for AI product builders |
|---|---|
| Cost per active user | How much does the platform spend serving typical usage? |
| Revenue per user | Does engagement translate into sufficient subscription, advertising or other income? |
| Inference efficiency | Can smaller models, caching or workload routing reduce costs without affecting quality? |
| Model strategy | Does the company need a proprietary foundation model, or can existing models support its differentiation? |
| Operational responsibilities | What will safety, privacy, support and regulatory compliance require? |
| Financial resilience | Can the company continue operating if additional financing becomes difficult? |
The answers will differ between products.
A business-focused AI application with high-value customers may have different economics from a consumer entertainment platform serving millions of free users.
Some applications may benefit from smaller specialized models. Others may require more capable systems despite their higher costs.
The appropriate technology strategy therefore depends partly on what customers are willing to pay and how much computing the product requires.
These are not questions that should be postponed until a platform reaches millions of users.
They influence product architecture, pricing and operational planning from the beginning.
A popular AI product is not necessarily a sustainable business
Character.AI developed a popular conversational platform, attracted millions of users and demonstrated substantial interest in AI-based entertainment.
Its revenue increased, and the company took steps to reduce computing expenses.
Yet it entered bankruptcy after facing financial and operational pressures that its existing business arrangements could not resolve.
That does not establish that consumer AI products are inherently unsustainable.
It demonstrates that adoption, technical capability and commercial sustainability are different measures of success.
For founders and technology professionals, Character.AI offers a case in which decisions about model architecture, infrastructure costs, monetization and user safety became closely connected.
Building an AI product that people want to use is a significant achievement. Building a business that can afford to keep providing it is a separate challenge.
Sources and further reading
Editorial source cutoff: 14 October 2026. The bankruptcy proceedings are ongoing, so figures, sale terms and timelines may change. The financial figures come from reporting on the CEO’s sworn declaration.
- Character.AI: Official Chapter 11 Announcement. October 12, 2026. Bankruptcy filing, operating status and sale process.
- Reuters: Character.AI Bankruptcy and Disney Transaction. October 13, 2026. Independent reporting on the bankruptcy and proposed licensing transaction.
- PPC Land: Character.AI Financial Disclosures. October 13, 2026. Analysis of the CEO’s sworn declaration, including revenue, infrastructure costs and cash burn.
- Alphabet: Form 10-Q for the quarter ended September 30, 2024. Primary documentation of the Google licensing transaction (Note 7).
- Character.AI: First 60 Days Update. August 2025. Model-development strategy following the Google agreement.
- Character.AI: Post-training image models for fandom. September 2026. Model adaptation and image-generation technology.
- Character.AI: April Update: New Model, Memory, and Lorebook. April 2026. Advertising, usage restrictions and product sustainability.
- The Verge: Now that Character.AI is going bankrupt, what happens to all of its users’ data?. October 2026. Privacy and regulatory concerns surrounding the bankruptcy proceedings.