Google bought a dead airline's email for $10 million, and Mercor bid $7.5M

Court filings unsealed this week show Google won a bankruptcy auction for Spirit Airlines' internal corpus. The Register has the itemized manifest, corroborated by Axios, CNN Business, Bloomberg Law, and Skift:

Roughly 100 million emails. 500 million Microsoft Teams items. 17 million OneDrive files. 20.5 million SharePoint items. 30+ million recorded customer service calls. 600,000 ServiceNow tickets. 763,000 flight records. 787,452 aircraft-parts purchase records.

The losing bidder was Mercor, an AI training-data company, at $7.5 million. Customer and loyalty data covering 97.5 million passengers was explicitly excluded, and Google says a third party will strip remaining PII.

Two prices got established here. The clearing price for a mid-size company's complete operational corpus is eight figures. And there was a competitive market for it, which means this wasn't a Google idiosyncrasy. A dedicated training-data vendor thought it was worth $7.5M.

Think about what's actually in 500 million Teams messages. Not documents. Not polished corpora. Ops people arguing about a delayed aircraft at 2am. Managers explaining a policy three different ways to three different people. Someone walking a colleague through a booking-system edge case. That's a kind of text that essentially doesn't exist on the public web, because it's the tacit operational knowledge of running a business, and nobody publishes it.

Same week, 404 Media put an AirTag inside one volume of a ~1,000-book bulk order placed anonymously through the Biblio marketplace and tracked it to the VGT3 section of Amazon's LAS8 warehouse in northeast Las Vegas, where workers cut the spines off books to scan them faster. The section logo is a T-Rex holding an open book. Booksellers had suspected the anonymous price-insensitive bulk buyer for a while. Now it's confirmed. TechCrunch, Futurism, and Simon Willison all picked it up the same day.

Both stories say the same thing from opposite ends. Text that was never online and never AI-contaminated is now the scarce input, and labs are acquiring it through bankruptcy courts and marketplace intermediaries rather than crawlers.

The builder angle isn't abstract. Your company's Slack, Teams, and email history is an asset on your balance sheet whether you've priced it or not, and in a bankruptcy it's an asset of the estate. If you're a founder, your data retention policy is now a term sheet issue. If you've got an enterprise contract with a vendor, go read what happens to your data if that vendor gets acquired or liquidated. I hadn't thought about that clause seriously before this week either.

I don't know how this ends legally. The PII stripping is a promise from the buyer, not a court order I've seen text of, and the excluded passenger data was excluded by negotiation.