Spirit Airlines shut down for good in May, after its second Chapter 11 filing in as many years. Its aircraft and gate slots went to the highest bidders in a fairly ordinary bankruptcy scramble. JetBlue alone paid $58.5 million for 22 of Spirit’s LaGuardia slots. Its Microsoft Teams archive turned out to be worth something too, just not to another airline.
Google has agreed to pay $10 million for a slice of Spirit’s internal data, according to Bloomberg Law. That beats a $7.5 million backup offer from AI recruitment and data firm Mercor. The US Bankruptcy Court for the Southern District of New York is due to hear the sale on Wednesday. Google says it intends to use the material to improve its products and train its AI models.
Inside the archive: 500 million Teams chats and more
The scale of what’s changing hands is what makes this a UC&C story rather than a straightforward AI or aviation one. According to Axios, court documents put the archive at around 100 million emails and roughly 500 million Microsoft Teams chats and collaboration records. Alongside that sits 17 million OneDrive files, more than 20 million SharePoint items, and 516 code repositories holding about 30 million lines of custom software. The archive also includes pricing models, booking curves, crew schedules and HR records.
Google says it isn’t buying customer data. The deal excludes Spirit’s roughly 97.5 million passenger profiles and an estimated 50.2 million Free Spirit loyalty records. A third party will strip personally identifiable information from what remains before transfer, with Google covering the cost. Google has also agreed not to attempt to re-identify anyone in the sanitised data.
For a Microsoft 365 tenant, that distinction doesn’t change the basic shape of what’s happening. A company’s internal collaboration history, built up over years of day-to-day coordination, becomes a saleable AI training asset once the company stops existing.
What the deal means for Microsoft Teams data governance
Microsoft has spent much of the past year telling enterprise customers the opposite story about their Teams data. Microsoft promises that Commercial Data Protection stops Copilot from storing or using Teams prompts and responses to train models, and it has made the same promise about Teams voice and face enrollment data. IT and compliance teams have largely taken that framing at face value. It shapes how comfortable they are putting sensitive conversations through Teams and Copilot.
The Spirit sale sits outside that framework entirely. A bankruptcy court is treating a defunct company’s Teams archive as a disposable asset, selling it to the highest bidder. That bidder can strip out obvious identifiers and use the archive for AI training. Microsoft’s governance commitments don’t cover what happens to years of internal chats once the company that generated them stops existing. None of the sensitivity labels, retention policies or DLP rules built under the frameworks pushed at Ignite 2025 account for it.
Why enterprise data is becoming AI’s new prize
Michael Curry, president of the data modernisation business unit at Rocket Software, sees deals like this becoming far more common as frontier AI labs push deeper into enterprise territory.
Frontier models are getting closer and closer to businesses. There are real advantages to that: more industry and business-specific knowledge can make these models much better at agentic workflows. But there’s also a risk of disintermediation, and we’re already getting glimpses of what that could look like.
Curry points to what happened when Anthropic launched a Claude legal plug-in in February: Thomson Reuters stock dropped 16% and Wolters Kluwer fell 13% in the days that followed.
Curry argues the Spirit sale differs from earlier rounds of AI data collection in the kind of data it treats as a saleable business asset.
A lot of what is necessary to understand how a business actually operates isn’t publicly available. It’s inside the enterprise, embedded in years of data, communications, applications, workflows, and proprietary knowledge.
Deals like this give frontier models direct access to those processes, he says. That helps explain the growing enterprise interest in sovereign data and sovereign AI. Companies want the benefits of increasingly capable models without losing control of the data and knowledge that makes their businesses valuable.
Curry also flags a second, less visible risk building alongside deals like Spirit’s: the agentic systems businesses are now layering on top of these models.
As agents get better at executing business processes, more proprietary knowledge is being captured in how those agents work. Companies need to think about where that knowledge goes, and whether it can make its way back to the frontier model providers.
His advice for enterprise IT and data leaders: plan for it now, not after the fact.
Organisations should now be thinking about how they build data platforms and design their agentic systems, so they are positioned to reap the benefits of AI while protecting the data and proprietary knowledge they’ve spent decades building.
A wider market for collaboration data from failed companies
Spirit isn’t the first case of this kind, just the largest by a wide margin. Reports earlier this year described failed startups selling old Slack chats and email archives to AI firms looking to extract value from collapsed businesses. Mercor, the runner-up bidder here, is already active in that market, offering individuals cash for materials from former employers. The driver is the same one pushing AI labs toward enterprise data generally. AI labs have largely exhausted publicly available web text. Internal workplace communications show how people actually coordinate, make decisions and fix mistakes in ways scraped web content doesn’t.
If bankruptcy courts keep treating Teams and email archives as saleable estate assets, Spirit won’t be the last case like this. Any organisation running Microsoft 365 should assume its collaboration data doesn’t necessarily disappear when the business does.
