Google buys 100 million Spirit Airlines emails and 500 million teams chats for $10 million to train AI models

Google is paying $10 million for something few people would expect to find at a bankrupt airline: roughly 100 million employee emails, 500 million Microsoft Teams messages, and years of internal business records that could help train its AI models.

The Alphabet-owned company won an auction for a vast collection of internal data belonging to Spirit Airlines, according to Business Insider. The trove includes employee emails, Teams conversations, spreadsheets, calendars, marketing materials, productivity data, and operational records. Some of the material reportedly stretches back decades.

Google Buys Spirit Airlines’ Data Archive for $10 Million in Bankruptcy Auction

Abhijay Rana, who first shared the deal on X, highlighted the unusual scale of the purchase: “Wow, seems like Google is buying Spirit Airlines’ enterprise data for $10m (outbidding Mercor at $7.5m). Basically includes every internal document, email, workflow, and codebase for a once $6B company.”

Google beat a $7.5 million offer from Mercor, an AI data company, for the collection.

The deal still needs court approval. A U.S. bankruptcy judge is expected to consider the sale at a hearing on Wednesday.

Spirit’s customer information isn’t part of the package. A third party will de-identify the data before transferring it to Google, removing personally identifiable information. That distinction matters, but the unusual transaction likely raises another question: How valuable have decades of corporate communications become in the AI era?

Google says it plans to use the Spirit data for product development and AI model training.

Corporate data could become AI’s next prized asset

For years, AI companies built increasingly capable models using enormous quantities of material gathered from the public internet, licensed datasets, books, code repositories, and other sources. Corporate records offer something different.

An airline’s internal communications capture how people actually work.

Emails show how employees discuss problems, make decisions, request information, and coordinate projects. Teams conversations contain shorter, conversational workplace exchanges. Calendars reveal how work gets organized. Spreadsheets contain structured business information. Operational documents can show how companies respond to real problems involving schedules, logistics, staffing, customers, and day-to-day operations.

Put together, hundreds of millions of these records could provide a detailed picture of how work happens inside a large organization.

That could become particularly valuable as Google and its rivals build AI agents meant to perform workplace tasks rather than simply answer questions. Training an AI system to operate inside a business requires more than knowledge gathered from websites. Developers need examples of how humans communicate, collaborate, make decisions, and move work from one system to another.

Spirit’s bankruptcy created an unusual opportunity to acquire that kind of material at enormous scale.

The airline shut down operations in May after struggling with heavy debt and high fuel costs, leaving its assets to be sold through bankruptcy proceedings. Its internal data has now emerged as an asset in its own right.

The $10 million price is tiny by Big Tech standards. The implications could be much larger.

Bankruptcy estates have traditionally focused on aircraft, equipment, intellectual property, real estate, customer lists, and other assets they can sell to repay creditors. The AI boom could add another category to that list: years or decades of internal corporate knowledge.

If proprietary workplace data proves useful for training AI agents, failed companies may possess something AI developers increasingly want. Their emails, chats, spreadsheets, workflows, and operational histories could become training material unavailable anywhere on the public web.

The Spirit deal could offer an early glimpse of that emerging market.

For Google, $10 million buys hundreds of millions of examples of people doing real work inside a real company. For everyone else, the transaction raises a more uncomfortable question: Could a company’s internal digital history become one of its most valuable assets after the company itself is gone?