Cover | Graphic by Pencil News

It's disheartening: The first-generation AI unicorn is going bankrupt.

On October 12, Character.AI, once valued at $1 billion, filed for bankruptcy reorganization. Without a buyer, the platform may ultimately shut down.

Character.AI focused on companionship, boasting 12 million users who spent an average of two hours daily chatting with AI. In August this year, its monthly revenue even reached $5.3 million.

What dragged Character.AI down was computing costs—it burned through $8–9 million in cash each month, far exceeding revenue.

- 01 - Investors Profited Before Bankruptcy

Character.AI’s two founders were top AI scientists at Google.

One, Noam Shazeer, co-authored the 2017 Transformer paper *Attention Is All You Need*, which underpins today’s global phenomenon, ChatGPT.

The other, Daniel De Freitas, also hails from Google. Both worked on Google’s conversational models, but Google hesitated to release them publicly over safety concerns.

In 2021, they left Google to launch Character.AI.

The platform officially launched in September 2022. It was addictive: Users could create their own AI characters or chat with historical figures, anime protagonists, and movie stars.

At the time, ChatGPT hadn’t launched yet. Character.AI quickly went viral.

Within five months, users had generated over 2 billion chat messages—the second billion took just one month. In May 2023, its mobile app hit 1.7 million downloads in its first week, 99% organically.

Silicon Valley investors took notice.

In March 2023, Character.AI secured $150 million in funding led by top-tier VC firm a16z, valuing the company at $1 billion. Investors included Nat Friedman, Elad Gil, and SV Angel.

They saw an exciting possibility: AI could be more than a productivity tool—it could dominate leisure time like games and social media.

For a while, Character.AI delivered.

By 2024, it had over 200 million registered users and 20 million monthly active users—remarkable for a three-year-old startup.

But the tide turned quickly.

In August 2024, Google announced a technology licensing deal with Character.AI and hired its two founders and about 30 R&D staff in a deal worth $2.7 billion (18 billion yuan).

Google didn’t acquire Character.AI outright—it bought some technology and poached its most critical talent. A significant portion of the $2.7 billion went to buying out early investors, who all profited.

The acquisition altered Character.AI’s fate.

Google took the top large model researchers, forcing the company to pivot: Instead of training its own foundational models, it shifted to third-party and open-source models, focusing resources on AI companionship and interactive storytelling.

By then, ChatGPT and others had flooded the market. Character.AI’s early scarcity vanished.

Users began leaving. Monthly active users peaked at 20 million in 2024 but fell to about 12 million by its bankruptcy this year.

In 2025, Character.AI hired Karandeep Anand, a former Meta, Microsoft, and Brex executive, as CEO to drive commercialization. But by September, Anand left to become Disney’s CTO.

By late September, 66 of Character.AI’s 86 full-time employees had accepted job offers from Disney. The AI unicorn was reduced to a platform with millions of users.

- 02 - Computing Costs Drained the Company

In 2024, Character.AI’s annual revenue was just $7.4 million; in 2025, it grew to $23.6 million. By August this year, monthly revenue hit $5.3 million.

On the surface, solid growth—but another number tells a darker story.

In 2024, Character.AI spent $13.8 million monthly on average for computing power. In 2025, that remained around $10 million per month. In the first eight months of this year, it paid $56.1 million in computing fees, accounting for 45% of operating expenses.

Its 2025 annual revenue couldn’t even cover three months of computing bills.

To survive, the company slashed costs.

It switched to cheaper GPUs, compressed models, cut staff, reduced office expenses, and lowered monthly operating costs from $21.9 million in late 2024 to $13.3 million by August this year.

But it wasn’t enough.

By August, monthly revenue was $5.3 million against $13.3 million in expenses—an $8–9 million monthly loss.

The root issue: AI companionship is fundamentally different from traditional internet businesses.

Every user message requires backend model computation. Longer chats mean more messages and higher computing costs. The more addicted users are, the more the platform loses.

Vega, founder of Shuban (an AI companionship startup), who extensively used Character.AI, told Pencil News that sustainable monetization requires solving memory issues first. “Without memory, characters can’t build stable relationships—users just try it once.”

Character.AI later realized this.

In May, it launched Story Memory to let users save key experiences; in July, it introduced Lorebook to help AI remember characters, locations, and story backgrounds.

But these features increased computing costs.

Vega believes the commercial path for AI companionship should be: Build emotional connections through chat, accumulate relationships via long-term memory, and then monetize around those relationships with content, services, or even personal agents.

Like *Honor of Kings*: Players enjoy the game for free but pay for heroes, skins, and other virtual items. The developer sells not just playtime but users’ affection for characters and sustained engagement.

AI companionship could follow suit. Users might pay for voices, avatars, stories, virtual gifts, or have the character accompany them into other apps.

But deep relationships don’t guarantee profits—a puzzle Character.AI never solved.

- 03 - AI Companionship Faces a Brutal Culling

After Character.AI’s bankruptcy, an unexpected buyer emerged: Disney.

Disney paid $20 million for a perpetual, non-exclusive license to some of Character.AI’s IP and hired related technical staff. But it didn’t buy the app or its private chat data.

Ironically, a year earlier, Disney had sent Character.AI a cease-and-desist letter over unauthorized Disney, Marvel, Pixar, and *Star Wars* characters on its platform.

Back then, Character.AI needed those famous figures to attract users. Now, the IP owner was taking its technology.

This highlights another industry dilemma.

While Character.AI had user-created virtual characters, many popular ones weren’t owned by the platform. The more famous the character, the pricier its copyright.

Disney, however, lacks no characters—or fans.

Behind Mickey Mouse, Marvel superheroes, and Pixar characters lies a mature ecosystem of films, games, parks, and toys. By integrating AI interaction with these characters, Disney could develop new interactive entertainment products and monetize them through existing channels. Disney hasn’t disclosed specific plans for the licensed technology.

Vega told Pencil News that when Shuban approached IP holders, high-quality rights owners were extremely cautious about AI licensing. “Important IP isn’t easily licensed; unimportant IP is widely available.”

In other words, for AI companionship companies without proprietary IP, creating a viral, profitable virtual character is far harder than imagined.

As the industry commercializes, another type of company is gaining attention—like Character.AI’s rival, Chai AI.

Chai also offers AI character chats. CEO William Beauchamp recently revealed annual revenue of $120 million.

Chai learned from Character.AI’s mistakes and began abandoning low-revenue users.

Beauchamp gave an example: In some regions, users consumed $1 in computing power but generated just $0.10 in revenue. Chai decided to focus resources on users who could cover computing costs rather than chasing scale.

Meanwhile, safety and compliance are becoming unavoidable costs.

After Character.AI’s bankruptcy, on October 14, attorneys general from 40 U.S. states asked the bankruptcy court to protect users’ private chat records, especially those of minors, during asset sales.

For traditional internet companies, user data is often a top asset.

But for AI companionship firms, chats filled with private emotions, personal experiences, and even sensitive information could become legal liabilities buyers avoid.

Looking back, Character.AI’s fate is particularly poignant. It proved people would spend two hours daily chatting with AI—but never proved how to monetize those hours.

This article does not constitute investment advice.