Google wins Spirit Airlines' data trove at bankruptcy auction to feed its AI

When Spirit Airlines shut down for good in May 2026 and went into liquidation, its data became just another asset to be sold off. Google won it at auction for $10 million, outbidding Mercor, an AI training-data specialist whose interest signals exactly why the trove is valuable. The haul is enormous: roughly 100 million emails, 500 million Microsoft Teams items, tens of millions of OneDrive and SharePoint files, more than 30 million recorded customer-service calls, 15 million-plus chat transcripts, and granular operational records covering flights, crew pairings, fuel purchases, and parts orders.

Google says it bought the data to improve its AI services, and the shape of the collection fits a growing industry bet that narrow models trained on deep domain knowledge outperform general-purpose LLMs — here, potentially an aviation-operations model built on a defunct carrier’s entire institutional memory. The operational depth of the dataset, from 763,000 flight records to fuel slips, is the kind of specialized corpus that’s otherwise almost impossible to assemble.

The privacy angle is the obvious flashpoint. Court filings state the data was deidentified before sale, and Google has pledged to strip any personally identifiable information it later finds. But with 30 million recorded calls and nearly 14 million active customer email addresses in the mix, deidentification at that scale is notoriously leaky — and re-identification through model outputs is a well-documented failure mode. The transaction also sets an uncomfortable precedent: customer data handed over for booking a cheap flight can end up training a tech giant’s models years later, sold by a bankruptcy court with no say from the people it describes.