On October 12, 2026, Character Technologies, Inc. voluntarily filed for Chapter 11 bankruptcy protection with the U.S. Bankruptcy Court for the District of Delaware, seeking to sell all or substantially all of its assets. In the sale process disclosed the following day, Disney affiliate Disney Worldwide Services became the stalking horse bidder with $20 million, in exchange for a non-exclusive perpetual license to certain intellectual property and a portion of the company's litigation claims.
Twenty-six months earlier, in August 2024, Google signed a $2.7 billion deal with the company operating under the name Character.AI. From $2.7 billion to $20 million, only a little over two years separated the two.
Where did the $2.7 billion go? Why did a star company with roughly 12 million monthly active users, whose active users chatted an average of 120 minutes per day, reach the point of needing a court-supervised sale? Why did 28 potential buyers produce only one unexecutable proposal, leaving Disney as the only willing bidder in the end? The answer splits in two: what Google bought was the people and the capability, not the company's future; what the company then faced on its own was a ledger where revenue could not catch up to inference costs, and a layer of legal risk that could not be priced.
What the $2.7 billion bought was the people, not the company's future
The structure of the August 2024 deal determined the nature of that money. Google's official line was that it obtained a non-exclusive license to Character.AI's generative AI technology. But the agreement also contained a key personnel clause: Noam Shazeer, Daniel De Freitas, and several employees returned to Google. The deployability of the consideration funds under the agreement was not disclosed.
Both Shazeer and De Freitas came from Google. The two had previously worked together on the LaMDA conversational AI project. Shazeer joined Google in 2000 and was one of the co-authors of the 2017 paper "Attention Is All You Need." They founded Character.AI in 2021, and the platform launched in 2022. These people were the real subject of the deal.
How high was that price? In March 2023, Andreessen Horowitz led a $150 million Series A round, valuing Character.AI at roughly $1 billion post-money. The Wall Street Journal reported that Google viewed Shazeer's return as the main reason for paying the high $2.7 billion price, a figure more than half the valuation Character.AI had previously sought from investors. The valuation the company asked the market for was less than twice what Google paid, and what Google wanted was mainly the people.
After the deal closed, the two sides maintained an independent posture for a time. A Google spokesperson denied holding equity in Character.AI, saying the other party remained an independent company. But the U.S. Department of Justice is investigating whether the $2.7 billion deal constitutes a de facto merger or acquisition and whether it circumvented antitrust review. In reporting from May 2025, the investigation was still at an early stage, and Google had not been formally accused.
The talent drain did not stop in 2024. In June 2026, Shazeer announced he was leaving Google to join OpenAI; De Freitas was still within the Google system at the time the related reports were published. Some Silicon Valley commentators argue that acqui-hire-type deals let tech giants "disarm" startups before they become threats. After the founding team left, the company still had to bear enormous costs on its own.
A chat that burns $8 million a month
At the time of the bankruptcy filing, the platform had about 12 million global monthly active users, and active users chatted an average of about 120 minutes per day. Over a month, that averages out to about 60 hours spent on the platform per active user. Both user scale and stickiness were considerable.
Revenue was another matter. Total revenue was about $7.4 million in fiscal 2024 and about $23.6 million in fiscal 2025, reaching about $5.3 million in the single month of August 2026. Of that, subscriptions accounted for about $2.8 million and advertising about $2.5 million. The subscription business grew from about $600,000 per month in January 2024 to $2.8 million per month in August 2026, with c.ai+ priced at $9.99 per month. The advertising business was only launched in mid-2025, generating about $500,000 in its first full month in July 2025 and rising to $2.5 million more than a year later, about 47% of that month's revenue.
Costs were on the other side. From January to August 2026, compute spending was about $56.1 million, about 45% of operating expenses, averaging about $7 million per month. Personnel spending was about $19.7 million, about 16%. Put the two sides together: in August 2026, the company's revenue was about $5.3 million, while operating expenses that month were $13.3 million.
The company did make real efforts to cut costs. Average monthly operating expenses fell from $21.9 million in the second half of 2024 to $13.3 million in August 2026, a decline of about 39%. Over the same period, average monthly compute fell from $13.8 million to about $7 million to $10 million. CEO Colin Adams listed the measures in his bankruptcy court declaration: no longer developing foundation models in-house, shifting inference workloads to lower-cost GPU capacity, and post-training open-weight models. The company relied more on external models and compute.
The math still did not add up. According to PANews calculations, revenue in August 2026 was about $5.3 million, while the company still lost about $8 million to $9 million that month, with revenue covering only 59% to 66% of the loss, leaving a monthly funding gap of about $2.7 million to $3.7 million. Net cash burn fell from an average of $21.5 million per month in the second half of 2024 to about $8.2 million in August 2026. Losses were narrowing, but always negative. As of the bankruptcy filing date, the company had about $57.414 million in unrestricted cash and about $59.414 million in cash and cash equivalents. Adams acknowledged that the company's technology operating costs were high and that it had long struggled with profitability. A $9.99 monthly fee and one round of advertising cannot cover the compute consumed by a single two-hour conversation.
28 potential buyers, one unexecutable proposal
If it were only about burning money, things might not have reached bankruptcy. What truly made buyers collectively balk was law and regulation.
On February 28, 2024, a 14-year-old boy in Florida, Sewell Setzer III, died by suicide. His mother, Megan Garcia, subsequently sued Character.AI, alleging the platform caused her son's addiction and deep immersion. This lawsuit became the starting point of the entire legal chain. In May 2025, a federal judge allowed Garcia's lawsuit to proceed, rejecting the company's First Amendment defense. Other families subsequently filed suits one after another.
By 2026, the litigation was no longer limited to private parties. On January 8, Google and Character.AI reached an "agreement in principle" over multiple family lawsuits, with terms undisclosed and the document containing no admission of liability. Reports said the company had settled five family lawsuits across four states before bankruptcy. On the same day, the Kentucky Attorney General sued Character Technologies, Shazeer, and De Freitas, alleging that the chatbot encouraged self-harm, starvation, suicide, and inappropriate sexual interaction, and violated the state's consumer data protection law. On May 1, Pennsylvania sued, alleging that the chatbot impersonated a licensed psychiatrist and provided a fake license number. Earlier, in September 2025, the FTC had ordered seven companies, including Character Technologies, to explain their child safety measures.
The scale of exposure can be represented by one number. According to PANews calculations based on data in the Kentucky complaint, as of March 13, 2025, there were more than 21.6 million U.S. independent accounts, of which more than 2.1 million were accounts for those under 18, with minors accounting for at least about 9.7%. The company's response was first to add mental health pop-ups and ban users under 18 from multi-turn chats, then on October 29, 2025, to announce a complete ban on minors using the chatbot. In September 2025, Disney had also sent a cease-and-desist letter, saying chatbots based on its characters were "harmful and dangerous to children," and the company said it had removed the relevant characters.
These pressures ultimately fed through to the negotiating table. The company began looking for buyers more than a year before bankruptcy, hiring an investment bank in August 2025 to contact about 28 potential buyers. The result was only one proposal, which was ultimately deemed unexecutable. Buyer feedback recorded by Houlihan Lokey was highly consistent: the product and user engagement were impressive, but litigation and regulatory pressure were too great, and they were unwilling to take it on. The founders were not able to stay out of it either. As of October 12, 2026, Shazeer and De Freitas were still listed as co-defendants in the Kentucky lawsuit. On October 7, the court had just granted an extension for them to submit information related to a motion to dismiss, and a response had not yet been filed.
Bankruptcy is not shutting downWhen it comes to bankruptcy, most people's first reaction is that the company is shutting down tomorrow. It is not. Chapter 11 is a court-supervised reorganization or sale process, not a liquidation.
The platform is still running for now. Paid subscriptions remain valid, and billing is unchanged for the time being. If service changes affect paid features, the company says it will notify subscribers in advance. Users can delete their accounts or export their data. The first hearing is scheduled for October 14, 2026, at 1:30 p.m. Eastern Time, before Judge Brendan L. Shannon. The company expects the sale and closing to be completed in late November or early December 2026, with an outside closing deadline of December 17, 2026. If a higher bidder willing to continue operating the platform emerges, the platform may be retained.
What does Disney's $20 million buy? A non-exclusive, perpetual license to certain intellectual property and some of the company's litigation claims. Disney has prepaid $7.5 million, of which $2 million is a refundable good-faith deposit. This price is subject to higher or better offers; it is the baseline bid for the auction, not the final sale price.
Three numbers need to be looked at together: $20 million, the $1 billion post-money valuation from the 2023 Series A round, and the 2024 Google deal worth $2.7 billion. According to PANews calculations, Disney's bid is about 0.2% of the Series A valuation and about 0.7% of the Google deal. But the three are not measured on the same basis: $1 billion is an equity valuation, $2.7 billion is consideration for technology licensing plus talent, and $20 million is a non-exclusive license and litigation claims in a bankruptcy sale. One cannot simply divide them to derive a decline. The bankruptcy filing estimates assets and liabilities both between $50 million and $100 million, and Law360 summarizes its debt as exceeding $50 million.
The company's official explanation for the bankruptcy is straightforward: the high cost of developing and expanding in the AI entertainment sector, combined with potential litigation liability and regulatory response, made it impossible to generate enough cash flow and difficult to continue operating under its existing business model. The people had already moved first. As of the end of September 2026, 66 of the company's 86 full-time employees had accepted job offers from Disney, and only 8 are expected to remain until the case concludes. Disney first hired most of the employees, then "leased" them back to the company for up to five months through a transition services agreement, charging only actual costs. Former CEO Karandeep Anand also joined Disney in September 2026 as senior executive vice president and chief technology officer.
Two details are worth noting. First, Google is the largest of the top 20 unsecured claims, at about $4.15 million. Second, Character.AI's privacy policy allows disclosure of chat logs, voice recordings, images, and videos when selling company assets or during negotiations, but the company has not said whether user chat logs are among the assets it plans to sell.
Back to the question at the beginning. Google's $2.7 billion bought the founding team and core model capabilities. After that, the company stayed alive on open-source models and external compute, and revenue growth could not keep up with inference costs. Combined with lawsuits from minor users and regulatory pressure, 28 potential buyers collectively balked. Bankruptcy protection and the $20 million floor price are the result of these two structural problems piling up.
404 Media commented that this is one of the first dominoes to fall in the long-predicted AI bubble: burning large amounts of money to run products, low revenue, high legal costs from user harm, and cost-cutting measures such as simplifying models, laying off staff, and injecting ads into chats did not work.
Before the December 17 outside closing deadline, there is still a possibility that a higher bid will emerge. This company's final pricing has not yet been hammered down.