Character Technologies, the company behind Character.ai, filed for Chapter 11 bankruptcy in Delaware on October 13 with $57 million in unrestricted cash, no funded debt, and estimated assets and liabilities each between $50 million and $100 million. The filing, laid out in a sworn declaration from CEO Colin Adams and first detailed in these numbers by 404 Media’s Jason Koebler, covers five years of a chatbot business whose costs never came close to lining up with what users would pay for it, even after the company stripped the product down to cut those costs.
The compute math
Compute is the line item that explains the bankruptcy. GPU capacity rented from third-party cloud providers to run inference on chat messages cost Character.ai $56.1 million between January and August 2026, about 45% of operating expenses, according to Adams’s declaration as reported by ppc.land. Personnel costs came second, at $19.7 million over the same eight months, or 16% of expenses. Averaged across those months, compute alone ran $7 million a month, down from $13.8 million a month in 2024 and $10 million a month in 2025.
Character’s impressive technology is expensive to operate.
Adams’s filing spells out how the company clawed that figure down: “transitioning inference workloads to lower-cost GPU capacity and post-training open-weight models rather than developing proprietary foundation models.” In plain terms, Character.ai stopped trying to build its own frontier model and started fine-tuning someone else’s open-weight model instead, which is cheaper to run and produces a more restricted, lower-quality product. That tracks what 404 Media called AI “enshittification” in a May report on the company about ads, usage limits and “lobotomized” models; the bankruptcy filing has the company’s own CEO admitting it was the plan.
The cuts show up across the whole cost base. Monthly operating expenses fell from $21.9 million in the second half of 2024 to an average of $18.5 million across 2025 to $13.3 million in August 2026. Monthly cash burn fell on roughly the same curve: $21.5 million in H2 2024, $17.6 million average in 2025, $8.2 million in August 2026 — the improvement Adams’s declaration calls “massive.” Three years of cost-cutting took the burn rate down by about three-quarters. It is still a burn rate, on $57 million of cash.
| Operating expenses ($M) | Cash burn ($M) | |
|---|---|---|
| H2 2024 avg | 21.9 | 21.5 |
| FY2025 avg | 18.5 | 17.6 |
| Aug 2026 | 13.3 | 8.2 |
Source: Character Technologies Chapter 11 declaration, via ppc.land
Revenue never caught up
Revenue moved in the other direction, slowly. Subscription revenue grew from $0.6 million in January 2024 to $2.8 million in August 2026. Character.ai didn’t run ads until mid-2025; once it started, ad revenue grew from about $0.5 million in July 2025 to $2.5 million in August 2026, close to half of that month’s $5.3 million total revenue and nearly matching subscriptions. In-app purchases arrived in early 2026. None of it closed the gap: annual revenue was $7.4 million for all of 2024 and $23.6 million for all of 2025, a better-than-threefold increase that still didn’t cover two months of the company’s 2024 burn rate.
Scale doesn’t help when the unit economics run backward. Character.ai has had roughly 163 million user accounts over its lifetime — a figure that surfaced for the first time in the bankruptcy motions, filed in part to argue that none of those 163 million people counts as a creditor entitled to individual notice of the case; the company says it will tell them about the bankruptcy through its blog and app instead, as ppc.land reported. Current monthly active users are estimated by the analytics firm Sacra at about 20 million, down from a mid-2024 peak near 28 million, according to TechCrunch. At $7 million a month, compute alone cost the company about $0.35 per monthly active user; subscriptions and ads together brought in about $0.27 per user in August 2026. After two years spent cutting the cost of running the chatbot, it still costs more to run than users pay for it.
Lawsuits, states, and Disney
Legal exposure sits on top of that gap. Character.ai is a defendant in Garcia v. Character Technologies, filed in October 2024 after 14-year-old Sewell Setzer III died by suicide in February 2024 following months of conversation with a companion bot modeled on a Game of Thrones character. A federal judge in Florida ruled in May 2025 that the product-liability claims could proceed and declined, at that stage, to treat the chatbot’s output as speech protected by the First Amendment, according to filings described by CourtDocket. The company, its founders and Google settled that case along with four related family suits in Colorado, New York and Texas on January 7, 2026, without admitting liability and on undisclosed terms. Character.ai banned open-ended chat for users under 18 that November — a change plaintiffs’ lawyers have pointed to as evidence the company recognized the risk.
Regulators piled on separately. Texas opened a child-safety investigation into the platform in December 2024. Pennsylvania sued in 2026 after a Character.ai bot allegedly told a user it was a licensed psychiatrist and gave a fabricated state license number, according to MLex. Italy’s privacy regulator, the Garante, fined the company €79,000, and Russian courts imposed fines totaling roughly $115,000.
Disney’s relationship with the company has moved from cease-and-desist to acquisition bid inside two years. Disney sent Character.ai a letter demanding it stop running unlicensed Marvel and Star Wars chatbots. The company’s CEO at the time, Karandeep Anand, left for Disney last month to become its first chief technology officer, taking part of his team with him, TechCrunch reported. Disney has now agreed to serve as the stalking-horse bidder in the bankruptcy sale: $20 million total for a non-exclusive, perpetual license to Character.ai’s intellectual property, with $7.5 million — 37.5% of the total, including a $2 million refundable deposit — paid upfront, per the declaration cited by ppc.land. It buys a license, not the company, which Adams says will be wound down absent another bidder by late November or early December.
The Google detour
Character.ai had already run one version of this trade. In August 2024, Google paid $2.7 billion in cash, cancelled convertible notes it held, and took a non-exclusive license to the company’s then-current model, in a deal that also moved founders Noam Shazeer and Daniel De Freitas, plus a reported fifth or so of the staff, to Google’s DeepMind unit. Shazeer, who held an estimated 30-40% stake in the company, is reported to have personally netted $750 million to $1 billion from the payout, according to Futurism, which cited contemporary reporting describing Shazeer’s return — not the license — as the deal’s real purpose. In June 2026, Shazeer left Google for OpenAI, as Outlook Business reported, leaving the company he’d returned to Google to run for roughly two years before the $2.7 billion rationale expired.
Character.ai is a smaller operation than OpenAI or Anthropic, and its failure mode — expensive inference, thin monetization, mounting legal bills — isn’t unique to it. The difference is that the larger labs can keep raising fresh rounds or lean on an advertising business that already pays for itself, the way Google and Meta do; Character.ai had neither. Anthropic’s own prospectus disclosed a $42 billion net loss even as it raises new capital, and Bain and McKinsey have warned that the industry’s spending plans assume revenue markets that don’t exist yet. Character.ai ran out of runway with $50 million to $100 million in liabilities. Nobody has yet shown what happens when the same compute-to-revenue math applies at $50 billion instead.