Twenty million dollars. That is the opening price Disney agreed to pay for assets of a company that investors valued at about $1 billion in 2023. In AI terms, it is a rounding error on a single training run.
On October 12, Character Technologies, the company behind Character.AI, filed a voluntary Chapter 11 case in Delaware to sell all or substantially all of its assets. Reuters reported that Disney agreed to buy some of those assets for $20 million, as the stalking horse bidder, with the platform continuing to operate during the sale. This is not a story about a chatbot failing to find users. It is a story about what a company actually owns.
What Disney is actually buying
The detail matters. According to the filing summary reported by Unite.AI, Disney gets a non-exclusive, perpetual license to certain intellectual property and claims, plus transition services from former employees Disney has already hired. Non-exclusive means Disney is not even buying the technology outright. It is buying the right to use it, alongside anyone else who ends up with a license.
The people went first. In September, Character.AI announced that its CEO, Karandeep Anand, was joining Disney as its Chief Technology Officer, and that members of the technical team were expected to follow him. By the time the bankruptcy was filed, the most valuable asset had already walked across the street.
And this was the second time. In 2024, Google struck a licensing deal for Character.AI's technology, reported at around $2.7 billion, and the company's founders, Noam Shazeer and Daniel De Freitas, returned to Google. The company kept the product and the users. The deal took the founders and a license to the technology.
The license-and-hire pattern has a cost
The 2024 deal was one of several structured the same way in that period. A large company licenses a startup's technology, hires its leadership and key researchers, and leaves a smaller company behind to keep running the product.
For the investors, it returned capital. For the founders, it was a soft landing. For the company left behind, it was an amputation.
What Character.AI shows is the long tail of that structure. A consumer AI product without the people who built the models, competing against assistants backed by the largest companies on earth, has a very short list of options.
Build your own frontier capability, which costs billions. Rent it from a competitor, which compresses margins. Or find a buyer for whatever is left.
The filing cited significant legal, regulatory and operational headwinds, and the company faced serious scrutiny and lawsuits over how its product affected young users. Those pressures were real. But a company with its own core technology and its own leadership has more room to absorb them than one that has already sold its future in pieces.
The lesson for every company building on AI
Most businesses reading this are not building frontier models. That does not make the lesson irrelevant. It makes it more relevant, because almost every AI product today is a layer on top of something it does not own.
The question to ask is simple: if your model provider, your founding engineers or your largest partner disappeared tomorrow, what would remain? For a lot of AI startups and AI features inside larger companies, the honest answer is a user interface and a prompt library. Both can be rebuilt by a competent team in weeks.
I made the case a while back that the model is not your moat. Character.AI is the inverse lesson. Its moat was supposed to be engagement, and it had plenty: millions of people spending long sessions with its characters. Engagement turned out to be rented too, because it depended on a product the company could no longer improve faster than the market.
What actually stays yours
So what does a defensible AI business own? In my experience, three things survive when the model layer shifts.
The first is proprietary data that improves the product and that nobody else can collect. Not generic usage logs, but structured knowledge about a specific domain or customer base. The second is distribution you control, a direct relationship with customers that does not depend on an app store ranking or a viral moment. The third is operational knowledge: the people who understand why the system works, documented well enough that their departure is a loss, not an extinction.
Character.AI had users and attention. It no longer had the founders, the CTO or exclusive control of its technology. When the market got hard, there was not enough left to defend.
Read the deal terms, not the headline
There is a practical point for buyers here too. If your company relies on an AI vendor, look at its ownership structure and its deal history.
Has it licensed its core technology to a larger company? Has its leadership moved on? Is it renting its models from a competitor that could change the price or the terms?
None of those are reasons to avoid a vendor. They are reasons to keep an exit plan, to keep your own data portable, and to avoid building a critical workflow on a product whose owner might be a bankruptcy court by next quarter. The sale and closing are expected in late November or early December. If no buyer continues the platform, it will be wound down.
Valuations measure what investors hoped a company would own. Bankruptcy courts measure what it actually did.