ARTIFICIAL INTELLIGENCER — The rise and fall of an AI chatbot darling

By Krystal Hu Oct 14 (Reuters) — This week, analysts and IPO investors gathered for private briefings in downtown San Francisco to hear new financials from the largest company ever to attempt to go public: Anthropic. Investors’ questions centered on revenue growth. Private companies can lean on so-called annualized revenue run rate, an estimate based on short-term revenue — the same metric behind the recent confusion over OpenAI’s reported numbers — but public companies answer to GAAP revenue instead, a stricter standard. And once a company is in the public market, it faces public scrutiny in full: How, and when, will it actually turn a profit?

That scrutiny will soon extend beyond Wall Street insiders, too. Retail investors are waiting for their own chance to finally get a piece of the action. But at the same time, they want an answer to the question: Will this AI lab founded five years ago live up to the roughly $2 trillion valuation it claims to be worth? Follow our live coverage to find out what Anthropic is telling investors.

AI, an early consumer AI darling. Scroll on. Will you invest in Anthropic after they go public? Share your thoughts by emailing me or following me on LinkedIn.

Forward this newsletter to your friend who would benefit from weekly news and insights on AI. They can also subscribe here. AI was one of the hottest names in AI. This week, it filed for Chapter 11.

It’s a case study of how a consumer AI product fails: A company lost its research backbone and founders, lost its fundraising momentum, lost users, and never found a way to make the math work against the compute bill required to keep its users talking. AI was co-founded in late 2021 by Noam Shazeer, a co-author of the landmark Transformer paper that underpins modern AI, and by his former Google colleague Daniel De Freitas. The two had built a chatbot together inside Google, only to watch the company refuse to release it publicly. So they quit, in frustration, and built their own.

At first, it worked, offering fun and novelty. AI let people chat and roleplay with AI versions of celebrities, anime characters, or personas they invented themselves. By 2024, it had more than 200 million registered users, 20 million monthly active users, and a $1 billion valuation. Then came the moment that offered rewards to founders and investors.

7 billion to license the technology and to bring Shazeer, De Freitas, and about 30 researchers in-house. In what Silicon Valley called a “reverse acqui-hire”, it allowed the tech giant to siphon off the best of the company, its talent, while avoiding regulatory review that a traditional takeover might have triggered. AI, minus the founders and talent, stayed afloat at least for a while. It leaned on third-party models instead of training its own, and pivoted hard into entertainment: AI companions, interactive fiction, short-form microdramas.

With the user base growth stalled and the company unable to raise more outside capital, the bankruptcy filing shows how upside-down the economics were. 8 million a month on compute. 8 million, and sometimes dipped below $1 million. 3 million.

1 million on compute. That’s a business model gap that was never going to close. Chai AI, its rival in the same character play-and-companion category, has made a deliberate call to walk away from growth for its own sake in less profitable regions, where it only generated 10 cents for every dollar of compute they spent, CEO William Beauchamp told me.

The company now generates $120 million in annual revenue, and “made a big strategic push to minimize any user whose cost is greater than their revenue, and instead be laser-focused on delivering the very best experience for the users who, for every dollar of compute, they generate more than a dollar of revenue,” Beauchamp said. AI learned, was safety. AI has faced a wave of lawsuits alleging its chatbots contributed to self-harm and suicide among teenage users, and the company itself cited that legal exposure, alongside regulatory pressure, as a direct factor in its bankruptcy filing. AI wasn’t alone in betting on this category.

Meta also poured resources into AI companions — proactive chatbots that message users first without prompting, remember past conversations, and in some cases have even been caught flirting with minors, drawing regulatory scrutiny and a temporary rollback of teen access this year. Reuters earned a Pulitzer Prize for investigations revealing how Meta knowingly exposed users, including children, to harmful AI chatbots and fraudulent advertisements. But Meta can absorb years of losses on an unprofitable bet, because advertising funds the rest of the company. AI never had that cushion.

By September, it was Disney’s turn to loot the talent. AI CEO Karandeep Anand to be Disney’s CTO. AI is now up for sale. Disney has placed the opening bid: $20 million for a non-exclusive, perpetual license to some of its intellectual property— less than 1% of what Google once paid.

If no better offer emerges by early December, the company says the platform will be wound down entirely, taking with it the private conversations of some 163 million user accounts, now sitting as assets in a bankruptcy estate. Ai Research To Read Does talking to an AI companion actually help with loneliness? For some people, it makes things worse, according to a recent Stanford research paper. AI users and collected full chat transcripts from 244 of them.

What they found: Just under 12% admitted companionship was their main reason for using the app, yet more than half described their chatbot as a “friend,” “companion,” or “romantic partner,” and over 80% of the donated chats centered on seeking emotional support. And among people with small real-world social circles, heavy AI companion use was linked to a lower sense of well-being. Oddly, the more personal information someone shared with their chatbot, the lower their well-being tended to be, the opposite of what self-disclosure usually does in human relationships.

The researchers think it comes down to reciprocity: A chatbot doesn’t share its own struggles back, yet it’s engineered to keep the conversation going regardless. They describe AI companionship as a “social snack” — a quick fix for loneliness that risks pulling vulnerable users further from the real connections that might actually help. (Reporting by Krystal Hu; Editing by Ken Li and Edmund Klamann)