Google is set to purchase internal data from the defunct Spirit Airlines to train its AI models.
A court document filed by PJT Partners, financial adviser to Spirit Airlines, specifies that Google will purchase 100 million emails and 500 million Teams chats as part of the deal.
It will also acquire hundreds of millions more pieces of data from OneDrive, SharePoint, shared calendars, spreadsheets, and many of Spirit’s internal business systems. Information on irregular operations, including all delays, recoveries, and logs surrounding disrupted flights, accounts for more than three billion data points alone.
The terms of the deal make clear that all data used by Google will be “deidentified” by a third-party organisation to remove all personal identifiers at Google’s expense.
Spirit Airlines ceased operations in May 2026 after a proposed government bailout fell through. It is now headed towards chapter 7 bankruptcy, under which all its assets will be sold.
Google initially bid $5 million and then raised this to beat a $7.5 million bid by Mercor, a US AI startup that connects experts with AI labs. The purchase awaits approval by a US bankruptcy judge, which is expected to be obtained on 19 August.
“We acquired part of an enterprise dataset from Spirit Airlines, which can be helpful in improving our products and AI models,” a Google spokesperson told Axios.
“We will not receive any personal information from this dataset. Any data we receive will be rigorously scrubbed of any personally identifiable information by a third party before receipt.”
Google has not specified whether it will use the data for training its flagship AI model Gemini, the open weight model family Gemma, or an as-yet-unannounced model.
AI labs seek out a diverse range of training data for their training runs to provide as much real-world context as possible. This primarily involves mass data scraping from the open internet, including copyrighted content, but business data is particularly prized as it helps to ensure AI models can carry out specialised tasks for enterprise customers.
