Data Marketplaces & Intermediaries

Google wins Spirit Airlines' data in bankruptcy auction for $10M, outbidding Mercor

Source: Skift · Aug 17, 2026

When Spirit Airlines filed for its second bankruptcy this year, its physical assets weren't the only thing on the auction block. Buried in the liquidation was a trove of internal business data — internal communications, financial records, and years of day-to-day operational documentation — and Google walked away with it for $10 million.

What makes this sale notable isn't the price tag, which is modest by big-tech standards. It's who else showed up to bid. Mercor, a company known in our space for AI training-data services and technical assessments, placed the second-highest bid at $7.5 million and lost. That's a meaningful signal: Mercor isn't just brokering or facilitating training data anymore — it's competing directly against a frontier AI lab to acquire raw enterprise data outright, at a price point that suggests real conviction about the asset's training value.

The dataset itself is substantial. Court filings describe roughly 100 million emails and around 500 million messages from Spirit's internal Microsoft Teams environment, layered with years of operational, finance, and payroll records. Notably, the sale explicitly carved out customer-facing personal information — this is a corpus of internal business communication and operational reasoning, not traveler PII. That distinction matters for how it can legally be used to train models: it's the kind of structured, real-world professional dialogue and decision-making trail that's genuinely scarce and hard to synthesize.

For anyone tracking the AI training-data economy, bankruptcy auctions are becoming an unlikely but increasingly important sourcing channel. Distressed companies sitting on years of internal communications and operational records are, whether they intended to be or not, sitting on assets that frontier labs now actively compete for. It also reinforces a theme we're seeing across the market: the highest-value training data increasingly isn't scraped from the open web — it's licensed, acquired, or auctioned from real organizations, with real provenance and real legal chain-of-custody, because that's what labs need to defend the data's use.

Key Points

• Google paid $10M for Spirit Airlines' business data in its bankruptcy sale
• Mercor was the next-highest bidder at $7.5M
• The dataset spans ~100M emails and ~500M internal Teams messages, plus operations, finance, and payroll records
• Customer-facing personal data was explicitly excluded from the sale
• Signals that AI labs are now bidding directly against assessment/data companies for bulk enterprise data