Summary of Key Points
The era of AI has given rise to a new business opportunity: acquiring the internal data of bankrupt companies—emails, chat records, project documents, etc.—and packaging it for sale to AI firms to use in model training. After the collapse of Spirit Airlines, a low-cost airline in the United States, its 600 million pieces of employee emails and Teams messages were purchased by Google for $10 million (about 1.67 cents per piece), potentially marking the first public auction of internal data from a bankrupt company. This phenomenon is driven by the increasing number of bankrupt companies and the demand for non-public data in AI systems. Additionally, corporate AI assistants (agents) need this type of real-world work process data, turning what was once considered digital waste into tradable assets. However, this industry is still in its infancy, with new participants, unclear pricing guidelines, and privacy concerns.
Why Have the Internal Chat Records of Failed Companies Suddenly Become Valuable?
In the past, when a company went bankrupt, only tangible assets such as aircraft, furniture, and trademarks could be sold; internal emails and chats were typically discarded. The reason for their current value is twofold:
1. Increasing number of bankrupt companies: In 2024, the failure rate of startups in the U.S. increased by 58%, with venture capital flowing heavily into AI (AI startups received $97 billion in funding), leading to many companies that relied on financing going out of business prematurely. For example, the financial technology company Tally, despite raising $172 million, still had to close down, leaving behind a wealth of internal data.
2. **AI's need for "real-world process data": Large AI models require massive amounts of high-quality data, but the quality content on public internet sources is dwindling (Epoch AI predicts it will be exhausted by around 2026). AI companies are seeking non-public data that records real work processes—how requests are made, how teams discuss issues, and how problems are resolved, which is not available on public websites. Corporate AI assistants particularly need this type of data to learn “how to do things” rather than just knowing “what to do.”
Who Is Selling the Data of Failed Companies?
In the past, only hardware and trademarks were sold during company liquidations. Now, there are three new types of entities involved:
1. Liquidation service providers: Such as SimpleClosure, which helps startups close down in a dignified manner. In 2025, it handled the closure of over 1,000 companies and launched an Asset Hub to sell their internal data (Slack records, Jira tickets, etc.). For instance, the transcription company cielo24 sold its three years’ worth of internal data through SimpleClosure for hundreds of thousands of dollars.
2. Data trading platforms: Companies like Protege, invested in by a16z, specialize in trading AI training data. They help find buyers for these datasets and have offered up to $300,000 per piece for some internal data.
3. Technical processing companies: Such as KLDiscovery, which convert emails and Teams messages into a format readable by AI systems. The successful auction of Spirit’s 600 million pieces of data was made possible thanks to their services.
How Much Do the Data of Failed Companies Worth?
There is no standard price; each transaction is negotiated individually:
- Small companies: Startup data sells for $10,000 to $100,000. For example, cielo24 sold its data for several hundred thousand dollars.
- Large companies: Spirit’s data was sold for $10 million (600 million pieces at 1.67 cents each). This price may seem low, but it represents additional revenue for the bankrupt company and unique process data for the AI firm.
- Comparison with other data: Photo licenses range from 5 cents to $1 per piece, and Reddit grants Google content licensing rights for $60 million annually. Spirit’s data is relatively cheap, but its value lies in the uniqueness of representing real corporate processes.
Pricing depends on factors such as the volume of data, industry (e.g., aviation data may be more valuable), time span, completeness, and the buyer’s purpose (data for training corporate agents is usually more expensive than for general models).
What Are the Steps Involved in Data Trading?
The process is not straightforward:
1. Data extraction: Messages from Slack and Microsoft 365 are exported in a usable format (e.g., JSON).
2. Anonymization: Personal information such as names and emails is removed, though not completely anonymized (possible identification through details like job titles and timestamps).
3. Court approval: The sale of bankrupt company data requires judicial permission; Spirit’s transaction was delayed due to objections from the flight attendant union, which raised concerns about employee privacy.
4. Delivery: The cleaned data is handed over to the buyer for use in AI training.
In the past, bankrupt companies sold customer data (e.g., 23andMe). This time, Spirit sold internal employee data, and since there are no established legal guidelines, the process is more complex.
The Significance and Controversies of This Business
- Significance:
- Preserving corporate knowledge: Previously, when companies failed, the experience gained by employees was lost. Now, this data can be preserved, allowing AI to learn from real work processes, preventing future companies from making the same mistakes.
- Additional revenue for bankrupt companies: What was once discarded waste becomes an asset that can help repay creditors.
- Controversies:
- Privacy risks: Anonymization may not be thorough enough, and employees’ chat records could still be identified by AI systems.
*Legal uncertainties:* Existing bankruptcy laws do not clearly regulate the sale of internal employee data, leading to concerns from unions and regulatory agencies.
- Ethical issues: Employees may feel that their chat records being sold without their consent violates their rights.
Spirit’s case serves as a precedent, highlighting the potential value of corporate digital heritage. It could futurely change how assets are listed during company liquidations and even affect employee contracts (e.g., by specifying data ownership in advance).
This industry is still developing, but it already shows that AI is not only transforming the operations of active companies but also redefining the aftermath of their failures. What was once discarded information can now be revitalized through data. How far this trend will go depends on the balance between privacy protection, legal regulations, and market demand.