The order book just updated. Google bought a dead airline's soul for $10 million. Not the planes. Not the routes. The emails. The Teams chats. The calendars. The booking records of 20,000 employees and millions of customers. Spirit Airlines, a bankrupt carrier that ceased operations in late 2024, just sold its internal business data—every Slack message, every HR file, every supply chain spreadsheet—to a tech giant for AI training.
Speed beats analysis when the graph is vertical. But here, the graph is the price of data assets, and it just went vertical. This is not a routine asset sale. This is a market signal that the AI data arms race has entered a new phase: bankrupt companies are now feeding the machine. I don't read whitepapers; I read order books. And this order book is full of raw, real-world business operations that no synthetic dataset can replicate.
Context: Why Now and Why Spirit
Spirit Airlines filed for Chapter 11 in late 2024. Its operations were grounded, planes returned to lessors, and employees laid off. The bankruptcy trustee needed to liquidate everything. Among the usual furniture and leases, a new asset appeared: a data package. The trustee reportedly listed it as "corporate data assets"—a vague term that turned out to be a goldmine.
Mercor, an AI data brokerage that specializes in sourcing training data for frontier models, bid $7.5 million. Google countered with $10 million and won. The spread of $2.5 million isn't just a premium; it's a strategic statement. Google wants exclusive access to this data, and it's willing to pay 33% above market to deny it to competitors.
But what exactly is in that package? According to published reports, the data includes:
- Internal emails and Teams chat logs (years of communication between departments)
- Calendar entries (meeting patterns, travel schedules, decision timestamps)
- HR records (performance reviews, salary data, employee relations)
- Customer data (frequent flyer profiles, booking history, preferences)
- Operational data (flight schedules, maintenance logs, supplier contracts)
- Marketing data (campaigns, conversion rates, customer segmentation)
This is not a random crawl of public internet. This is a high-fidelity recording of how a real airline operates—complete with the messiness of human decision-making, supply chain bottlenecks, and customer service crises. For any AI company trying to build enterprise agents, this is the holy grail.
Core: What Google Actually Bought
1. The Data Pipeline
Let’s talk about what Google will do with this data. First, anonymization. The Spirit statement says they will "remove personal identifiers" before transfer. But anonymization is a spectrum. At the low end, you strip explicit PII (names, social security numbers, email addresses). At the high end, you apply differential privacy and formal privacy guarantees.
Based on my experience reverse-engineering DeFi protocols in 2020, I know that the difference between a sanitized dataset and a usable one is the preservation of semantic relationships. If you delete the email addresses but keep the full text of internal communications, the model can still infer who is talking to whom based on context. Google's data pipeline will likely use a tiered approach: remove direct identifiers, apply k-anonymity on structured fields, and then use a transformer-based model to redact sensitive passages in unstructured text. The cost of this processing is non-trivial, but the value of the data justifies it.
2. The Training Use Case
This data is not for pretraining a general-purpose LLM. It's too narrow, too domain-specific. Google will use it for fine-tuning Gemini Enterprise, its AI agent for business workflows. Imagine an AI that has seen thousands of real airline scheduling conflicts, real customer complaints, real HR disputes. It learns the patterns of corporate decision-making, the jargon of operations, the unwritten rules of corporate culture.
More importantly, the data includes Teams chat logs. Microsoft Teams is the dominant enterprise collaboration tool. Google Workspace has its own tools (Chat, Meet, Calendar), but it lacks the massive corpus of real-world usage that Microsoft has. By acquiring Spirit's Teams data, Google gets a window into how people use a competitor's product—a form of competitive intelligence that is extremely hard to obtain otherwise.
3. The Market Signal
Mercor's bid of $7.5 million and Google's counter of $10 million establish a new pricing benchmark for bankrupt company data. This is not a one-off. The bankruptcy court will now see data as a salvageable asset. Expect more trustees to list "corporate data" as a line item, and more AI companies to bid.
I've seen this pattern before. In 2022, during the FTX collapse, I tracked VC whitelists and realized that data was the last asset to be valued—until it wasn't. The best news is the news that moves the price. The price of data assets just moved, and it's going to create a whole new market.
4. The Competitive Moats
Why did Google outbid Mercor? Mercor is a data broker; it would have sold access to the data to multiple AI labs. Google's acquisition is exclusive. This means OpenAI, Anthropic, and Microsoft cannot use this data to train their own enterprise agents. The $2.5 million premium is a classic moat-building strategy.
But there's a deeper angle. Google's Workspace competes directly with Microsoft 365. The Teams chat data is particularly valuable because it allows Google to train its AI to understand the nuances of Microsoft's collaboration tools. This is not just about data; it's about understanding the enemy's terrain.
Contrarian: The Unreported Blind Spots
Everyone is talking about the AI opportunity. Nobody is talking about the privacy landmine.
The data includes employee communications. These employees did not consent to have their internal messages sold to Google for AI training. Bankruptcy law allows the sale of assets, but privacy laws may override. Under GDPR, any processing of personal data requires a lawful basis. The bankruptcy sale may not qualify as "legitimate interest"—especially when the data is used for commercial AI training rather than the original business purpose.
Even if Google anonymizes the data, the risk of re-identification is significant. Internal emails are high-dimensional. A single email chain can contain references to specific projects, dates, and locations that, when combined, can identify individuals. The model might memorize and later regurgitate sensitive information—performance reviews, medical leave requests, internal conflicts. I've audited on-chain data leaks; this is a similar vector, but with higher stakes.
There's also a procedural blind spot. The bankruptcy court approved the sale without a public privacy impact assessment. The trustee did not notify employees that their data was being sold. This sets a dangerous precedent: every bankrupt company with a digital footprint becomes a target. Expect a wave of class-action lawsuits, especially in EU jurisdictions.
And let's not forget the ethical dimension. Google is buying the last remaining asset of a failed company: the memories of its employees. The phrase "data capital" has never been more literal.
Takeaway: What to Watch Next
The bankruptcy court is expected to approve the sale within the next week. If it does, the floodgates open. Every restructuring law firm will add "data monetization" to their checklist. Every AI company will start monitoring bankruptcy filings for data assets.
But the real action is regulatory. The EU's AI Act and GDPR are watching. If the European Data Protection Board issues a statement, this deal could be frozen. The US CFPB might also weigh in on consumer data sales.
My play? I'm not shorting Google. I'm long on data privacy litigation. The first class-action lawsuit will drop within 90 days. And when it does, the price of bankrupt data assets will either skyrocket (if the court upholds the sale) or crash (if regulators intervene).
Speed beats analysis when the graph is vertical. But the graph here is the regulatory mood, and it's about to spike. I'll be watching the docket, not the headlines.