Spirit Airlines spent two decades teaching Americans that a plane ticket could cost less than the bag going under the seat. The airline is gone. Its internal correspondence is now a Google training asset, and the clearing price was $10 million.
Google won the estate’s data in a bankruptcy auction, outbidding the AI recruitment startup Mercor by $2.5 million against a reported $7.5 million offer. CNN reported the sale, and the inventory is not a customer database. It is the operating memory of a functioning airline.
- Roughly 100 million employee emails
- About 500 million Microsoft Teams messages, plus calendars, documents and spreadsheets
- Around 30 million lines of program code, development data, and internal software models
- Pricing data covering 7.2 billion competing flights
- Approximately 7.5 billion passenger transactions
Forbes put the arithmetic plainly: a decade-plus of a major carrier’s decision-making, sold for roughly the cost of a Super Bowl ad. Tom’s Hardware, cataloguing the same trove, called out how little $10 million buys relative to what is in the box.
What Google Is Actually Buying
Frontier model training has a supply problem. The public internet has been scraped, the licensing deals with publishers are priced accordingly, and synthetic data has known limits for teaching a model how organizations behave. What almost nobody has is a complete, longitudinal, internally consistent record of a real company arguing with itself.
That is precisely what an email and Teams archive is. Not facts about airlines. Process. How a pricing analyst justifies a fare change, how an operations manager escalates a maintenance delay, how a marketing team fights with finance over a promotion, how those threads resolve and what happened next. Pair that correspondence with 7.2 billion competing-fare records and you can trace a decision from the argument that produced it to the revenue outcome it generated.
For a company building agents meant to perform white-collar work, that lineage is the product. Google has said the purchase supports product development and AI training, which is accurate and tells you very little.
The scarce asset in enterprise AI was never information about business. It was a complete record of a business thinking.
Mercor, which recruits domain experts to generate exactly this kind of professional reasoning data by hand, understood the value well enough to bid $7.5 million. It lost to a buyer that can absorb the whole estate at once. The efficient path to a decade of expert judgment turned out to be waiting for the expert’s employer to go under.
The Consent Nobody Gave
The privacy architecture here is real, and it is also narrower than the headlines suggest.
Court filings specify that Google does not receive personal data. The 97.5 million passenger profiles Spirit maintained are excluded, as are the 50.2 million customer records tied to the Free Spirit loyalty program. A court-appointed third-party ombudsman oversees deidentification, which obliges “reasonable measures” to strip material that could be linked to an individual before Google takes delivery. Judge Sean Lane still has to approve the transaction.
Notice the shape of those protections. The carve-outs are consumer records. What survives into Google’s hands, scrubbed of identifiers, is the employee correspondence, and the people who wrote those 100 million emails are not the party the consumer-privacy machinery was built to protect. They wrote to a colleague on a Tuesday about a delayed aircraft. They complained about a manager. They negotiated a vendor contract in a thread that ran for months. None of them consented to any of it, because there was no moment at which consent was a question anyone was asked.
Deidentification also does a specific job and only that job. It removes the name attached to a message. It does not remove the message, and at this corpus size the interesting signal was never the name. It was the pattern of reasoning across hundreds of thousands of threads, which survives anonymization intact because it has to. That is the entire reason the archive is worth buying.
Bankruptcy Is the Loophole, and It Is Wide Open
Spirit’s employees had a privacy relationship with Spirit. When the estate is liquidated, that relationship becomes an asset with a price, and the counterparty changes without anyone’s permission. The promise was made by a company that no longer exists, and the obligation does not reliably follow the data to whoever buys it.
There is no shortage of firms in a position to find this out. Every distressed retailer, every collapsed fintech, every logistics company that overextended has the same asset sitting on its servers, and until this week most creditors’ committees would have valued it near zero. A $10 million comparable changes that. The next bankruptcy adviser preparing an asset schedule now has a reason to list the Slack archive.
The pricing pressure runs the other way too. A creditor recovering pennies on the dollar has a fiduciary reason to sell, and an employee objecting on privacy grounds has no standing in that calculation and no lawyer in the room. The incentive structure produces one outcome.
Data retention policy is not an abstraction here. It is the only lever that determines how much exists to sell when a company fails. We watched a version of that fight play out when Flock Safety cut its retention window to seven days after city councils kept voting its contracts down. Retention limits look like compliance overhead right up until the moment an estate is being carved up, at which point they are the difference between a $10 million asset and nothing to sell.
Spirit’s competitors, meanwhile, should read the fare-data line twice. Pricing intelligence on 7.2 billion competing flights describes an entire industry’s behavior, and it just changed hands in a proceeding none of those airlines were party to.
The airline died of thin margins in a market it helped create. Its filing cabinets outlived it, and they turned out to be the profitable part.