Google has secured a massive data package from bankrupt US budget airline Spirit Airlines to train its AI models. For $10 million (approximately €8.6 million), the company won an auction for roughly 100 million emails and 500 million conversations on Microsoft Teams, according to documents filed in the airline's insolvency proceedings. The package also includes software and pricing data.
Key Facts
- Google paid $10 million for Spirit Airlines' internal communications data
- Dataset comprises 100 million emails and 500 million Teams chats
- Rival bidder Mercor (AI firm) offered $7.5 million but lost the auction
- Google pledges: All personally identifiable information will be removed before handover
Why Insolvency Data Matters
The purchase reflects a trend increasingly visible in Silicon Valley: AI companies are deliberately acquiring internal communications from failed startups and enterprises. The logic is straightforward – such data reveals genuine, day-to-day organizational workflows. This helps AI models learn realistic scenarios rather than just theoretical patterns.
Google itself emphasizes the value: the dataset can help "improve our products and AI models," the company stated in a comment to Bloomberg. The premise is that AI learns more from authentic business processes – from email traffic to decision-making – than from synthetic or publicly available text.
The Bidding War
Google was not alone in its interest. AI firm Mercor also participated in the auction but fell short with a $7.5 million bid. This demonstrates that the market for such data assets is real and competitive.
Implications for German Companies
For enterprises in Germany, this raises a dual question. First: what data could become AI training material during insolvency – and who benefits? Second: is anonymizing personally identifiable data sufficient when dealing with sensitive business communications? Google promises that a third party will handle anonymization, but the effectiveness of such measures remains unclear.
German insolvency administrators and creditors may face similar situations in the future. The question is not only legal (data protection, trade secrets) but also strategic: should data from failed firms be sold as raw material for AI training – or should stricter rules apply? So far, there is no clear answer.
Sources
Editorially owned by Ideal Syka. Sources and method: Newsroom & method. Tips and corrections: ai@i6eal.de.
