Character.AI Had Users For Hours A Day. It Still Went Bankrupt.

In 2023, Character.AI said its users were averaging more than two hours a day on the platform. Most consumer apps would kill for that number.

On October 13, 2026, the company filed for Chapter 11 bankruptcy.

It had raised more than $193 million from investors including a16z, Lightspeed, and Greycroft, according to PitchBook data cited by TechCrunch. It hit 20 million monthly active users in late 2024. Google paid $2.7 billion to license its models. And still, five years after founding, the court papers list assets and liabilities between $50 million and $100 million, with some assets heading to Disney for $20 million and no buyer yet for the rest.

There were lawsuits, regulatory probes, and leadership changes. Those mattered a lot. But TechCrunch pointed at something every AI founder should tape to their monitor: high engagement can be a double-edged sword, because AI costs rise the more people use the app.

Let's unpack that, because it's the most transferable lesson in the whole story.

A fast rise, a talent licensing deal, safety crises, and a pivot that didn't fix the economics.

Here's the short timeline from public reporting:

• Late 2021: Former Google engineers Noam Shazeer (a co-author of the Transformer paper) and Daniel De Freitas found the company.
• March 2023: A $150 million round led by a16z values it at $1 billion. The company says users exchanged over 2 billion messages in five months.
• May 2023: Its apps pass 1.7 million installs in the first week.
• August 2024: Google licenses its models in a $2.7 billion deal. Both founders return to Google. The company stops training foundation models and focuses on growth.
• October 2024: 20 million monthly active users, and a lawsuit tied to a teenager's death. Texas opens a child safety investigation soon after.
• 2025 to 2026: Settlements, new safety features, a ban on minors, a new CEO, and new products like AI video, social feeds, and microdramas.
• September 2026: Former CEO Karandeep Singh Anand and staff join Disney.
• October 13, 2026: Chapter 11, with a sale process expected to close in late November or early December.

The safety story deserves its own serious discussion. This piece is about the business model underneath it.

In traditional software, an extra hour of use costs almost nothing. In AI, every reply costs real compute.

Think about a photo app in 2015. Once the servers are running, a user scrolling for an extra hour barely moves your bill. So engagement is close to pure upside. You can sell ads against it and margins grow with scale.

AI chat flips that. Every message a user sends triggers model inference, which means GPUs doing work you pay for. A user chatting for two hours a day isn't a cheap fan. They're your most expensive customer.

That creates a squeeze. Your heaviest users cost the most to serve. If they're on a free plan, they drain cash fastest. If they pay a flat subscription, they can still cost more than they pay. And ads, the usual fix for free consumer apps, have to cover a much higher cost per session than they ever did for social feeds.

Character.AI ranked among the top apps for time spent, according to Sensor Tower. That's the kind of chart that looks fantastic in a pitch deck and scary in a cost model.

Build a per-user cost model that scales with usage, then check it against what each user pays.

You don't need a finance degree for this. You need four numbers per user segment:

Using those made-up numbers: 3 x 40 x $0.002 = $0.24 a day, or about $7.20 a month. Against $4 of revenue, you lose $3.20 on every user in that segment. Now double their engagement and your loss more than doubles.

Run it for your light, medium, and heavy users separately. Averages hide the problem. Your heaviest 10% often drive most of the cost.

You can do this in a Google Sheet, in Causal, or in a planning tool like Foundra that has templates for financial projections. What matters is that usage is an input in the model, not an afterthought.

Alt text: Chart of AI cost per user climbing above flat subscription revenue as engagement grows Caption: In AI products, the lines can cross. Find where before your users do.

Price on usage, route to cheaper models, and design for value per message rather than time spent.

Here are the options I see working, roughly from easiest to hardest:

1. Usage tiers. Free plans get a daily cap. Paid plans get more. Power users pay for what they use. It feels less generous, but it keeps you alive.
2. Model routing. Not every reply needs your best model. Send casual chatter to a small, cheap model and save the expensive one for hard tasks.
3. Caching and summarizing. Long conversations get expensive because context grows. Summarize old turns instead of resending everything.
4. Charge for outcomes. If your product finishes a job, like drafting a document, price the job. Then a long session is a feature, not a cost leak.
5. Rethink the north star. If your main metric is minutes per day, your team will chase the thing that costs you most. Consider "tasks completed" or "paid users retained" instead.

And be careful with "we'll monetize later." That line works for products where usage is nearly free. In AI, later can arrive faster than you planned.

Engagement-maximizing products attract scrutiny, and that scrutiny carries real costs.

It's impossible to separate Character.AI's finances from its legal problems. The company faced lawsuits tied to teen users, a state investigation, settlements, and eventually banned minors from the platform. CEO Colin Adams described "significant legal, regulatory, and operational headwinds" in his bankruptcy statement.

For founders, the lesson isn't only "be careful." It's that design choices driving extreme time-on-app can create legal exposure, force expensive safety work, and shrink your addressable market (banning minors removes users). Those costs belong in your plan from day one, especially if your product forms emotional relationships with users.

So ask yourself early: if a regulator, a parent, or a journalist looked at our engagement chart, would they see delight or dependency?

Pull last month's inference bill and divide it by active users. That one number tells you a lot.

Then split it by usage level. If your top 10% of users cost more than they pay, you've found your first pricing problem. Fix it while your user base is small, because changing plans on 20 people is a conversation and changing plans on 20 million is a crisis.

Also check your free tier. Is it a sample or a whole meal? A good free tier shows value fast and leaves people wanting more. A bad one gives away your most expensive feature forever.

Engagement is only good news when each extra hour makes you money.

• Character.AI filed for Chapter 11 on October 13, 2026, after raising over $193 million and reaching 20 million monthly users.
• In AI products, every reply has a compute cost, so heavy users can be your most expensive customers.
• Model cost per user by segment (light, medium, heavy), not on averages.
• Use usage tiers, model routing, context summarizing, and outcome pricing to keep costs below revenue.
• Pick a north star metric that doesn't reward the behavior that costs you most.
• Safety, legal, and regulatory work is a real line item for engagement-heavy consumer AI.

Why did Character.AI file for bankruptcy? The company cited legal, regulatory, and operational headwinds. Reporting also pointed to the difficulty of monetizing a consumer AI app where costs rise with usage. It filed Chapter 11 on October 13, 2026, and plans a court-supervised sale.

Is Character.AI shutting down? Not immediately. The company said it will keep operating during the sale process. If no acceptable bid arrives, it said the platform will be wound down.

What did Disney buy from Character.AI? Some assets for $20 million plus a non-exclusive perpetual license to its technology. Former CEO Karandeep Singh Anand and some staff joined Disney in September 2026.

Why are AI apps more expensive to run than normal apps? Each AI response requires model inference on GPUs, which costs money every time. Traditional apps have near-zero cost for extra usage once the servers are running.

How should I price a consumer AI product? Start with your per-user cost at different usage levels, then choose tiers or usage caps so your heaviest users still cover their costs. Test pricing early instead of waiting until you have scale.

Is high engagement bad for AI startups? Not by itself. It becomes a problem when revenue per user doesn't rise with usage. If each extra hour earns more than it costs, engagement is great.

1. Just 5 years in, interactive chatbot platform Character.AI files for bankruptcy, TechCrunch, October 13, 2026
2. Character.AI Secures $150M in Series A Funding Led by Andreessen Horowitz, Business Wire, March 23, 2023
3. Character.AI CEO Noam Shazeer returns to Google, TechCrunch, August 2, 2024
4. Disney's first CTO led an AI startup it once accused of copying its characters, TechCrunch, September 18, 2026
5. Character.AI Chapter 11 petition, Stretto case filing