Character.AI's bankruptcy filing admits the economics never worked

Share
Character.AI's bankruptcy filing admits the economics never worked

The filing that landed Monday said the company was out of money. The affidavit behind it, reported today, says it never had a path to the money in the first place — and puts numbers on a claim the whole AI industry has been arguing about for three years.

Character.AI's own chief executive told a bankruptcy court that compute cost the company roughly five times what its subscriptions brought in — and that "at all times, Character has struggled with profitability." According to the first-day affidavit from CEO Colin Adams that 404 Media reviewed, Character.AI spent $13.8 million a month on AI compute in 2024 and $10 million a month in 2025. Its subscription revenue peaked at $2.8 million a month and was sometimes under $1 million; revenue from all sources, ads included, topped out at $5.3 million a month. Between January and August 2026 it spent $56.1 million on compute and $19.7 million on salaries and benefits, and as of August it was still losing roughly $8 million to $9 million a month — a run rate Adams calls a massive improvement, arriving after the company cut headcount, moved inference onto cheaper GPUs, post-trained open-weight models instead of building its own, and injected ads and in-app purchases into the product.

The detail that matters for everyone else is how it cut the burn: selling the product down. 404 Media described the same strategy in May as AI enshittification — heavier ads, usage caps, guardrails, a "lobotomized" model — and the affidavit now confirms that was the plan, because it was the only lever a company with no other business could pull. Adams also concedes that legal costs from the harm its chatbots caused contributed to the collapse, and that the business was unsustainable even without them. The honest reading: Character.AI did not fail because it mismanaged a good business, it failed because the frontier labs can hand out better conversation for free — the affidavit simply prices the gap between compute and subscription revenue at roughly 5x. We covered the filing itself on Monday — Character.AI files for Chapter 11.


A federal judge handed Apple expedited access to OpenAI's interview records in the trade secrets suit, and ordered three former Apple staff deposed. Judge Edward Davila ruled that Apple can see interview checklists, guides, manuals, and scripts from OpenAI's hiring of ex-Apple employees into its consumer hardware and robotics groups, plus technical presentations that reference Apple technology — five days to hand over material already collected, fourteen days for the rest. OpenAI must also produce private LINE messages between employee Chang Liu and Alyssa Peng, and all three of Liu, Peng, and chief hardware officer Tang Yew Tan will be deposed. Apple didn't get everything: the judge denied expedited access to OpenAI's work on metal finishes, power management, and battery technology, and denied the broader requests around what ex-employees built — leaving the door open to narrow them and refile. Apple sued in July over what it describes as a pattern of theft.

What to watch: whether the second tier of consumer AI starts consolidating rather than raising, and whether Apple's discovery turns up anything that reaches the merits of its case.

If compute is the floor and subscriptions never clear it, can any consumer AI company survive without a bigger business behind it? Tell us in the comments.

Read more

Anthropic migrated off git after merges hit 30,000 a day

Anthropic migrated off git after merges hit 30,000 a day

The story under the story today is the infrastructure: what happens when agents write code faster than the tools built for humans, and what a lab does when the threat turns out to be inside the building. Anthropic says its merge rate grew more than tenfold this year, so it replaced git with a version control system it wrote itself — built and fully migrated by seven engineers in five months. In May 2026, the company averaged around 2,500 code changes merged per day across its monorepo; by Sept

Google Play opens a private lane for vibe-coded apps

Google Play opens a private lane for vibe-coded apps

Google Play now hosts vibe-coded apps that only you can use. Google announced Tuesday that you can build a personal Android app in Google AI Studio, then install it on your own phone straight through Play — no developer account, no public listing. The app never appears in Play's search or listings and is visible only to its creator. That closes a real gap: until now, putting your AI-built app on an Android device meant publishing it to Play's roughly 3 billion users, which is why most people who

Gemini 4 Argon hits general availability, free tiers left out

Gemini 4 Argon hits general availability, free tiers left out

Google spent months as the chase car in the frontier race — today it put its flagship model in buyers' hands, while one of Anthropic's founders moved up his own timeline for when the field starts improving itself. Google released Gemini 4 Argon to the public today — its first frontier model since Gemini 3, and the one it needs to stop trailing Anthropic and OpenAI. The rollout reaches developers through the AI Studio and Google Cloud APIs, Antigravity for Google AI Ultra subscribers, and the G

Anthropic to reveal first Q3 numbers at pre-IPO investor day

Anthropic to reveal first Q3 numbers at pre-IPO investor day

The private AI market's most important number lands Wednesday: Anthropic is handing institutional investors its third-quarter financials for the first time — and after last week's OpenAI revenue whiplash, every decimal will be read as a signal about the whole AI trade. Anthropic will publish third-quarter financials to institutional investors at a meeting at its San Francisco headquarters on Wednesday, according to two people familiar with the matter — the first time the lab's Q3 numbers have