Key Takeaways

  • Google paid $10 million for Spirit Airlines' enterprise data during bankruptcy proceedings, leaving the airplanes behind.
  • Merkor also reportedly competed in the bankruptcy bidding process for the same corporate dataset.
  • Real transactional and operational history cannot be generated synthetically, making legacy corporate archives a primary target for training business agents.
  • Model architectures and raw compute are commoditizing quickly, turning proprietary historical datasets into the main durable moat for software companies.

The New Liquidation Target

When Spirit Airlines entered bankruptcy, the most contested asset on the auction block was not a fleet of passenger jets or airport gate leases. It was raw operational records. Google paid $10 million to acquire the carrier's proprietary data. Rumors during the bankruptcy process pointed to Merkor as another bidder competing for the same records.

As Ofir Ehrlich explained, “Google buy a something from the bankrupt Spirit Airlines. They didn't buy airplanes. They bought the data. They bought the data for $10 million because they think it's very important in that perspective.”

This auction shows how AI development priorities have shifted. Years ago, technology companies scraped public websites for text and code. Today, foundation models and compute clusters have low switching costs. Anyone can rent compute or call an API. The scarce resource is authentic operational history: passenger booking patterns, pricing adjustments during disruptions, flight cancellations, and customer service ticket resolutions.

Why Synthetic Data Fails for Business Agents

Building autonomous agents that run actual business workflows requires messy, authentic corporate records. You cannot prompt an LLM to invent years of accurate discount pricing behavior under jet fuel price swings. Synthetic data creates smooth, idealized patterns. Real businesses run on edge cases, human errors, and messy trade-offs.

Gonen Stein pointed out that Google already operates in travel search and booking: “Google obviously is in this travel space for a while. They want this type of data. They're already monetizing it. This allows them to understand, train it, understand it, monetize it even further.”

AI labs are now actively hunting through legacy corporate archives, hedge fund records, and distressed assets. They need training data that captures how businesses actually transact. As Ehrlich observed, tech data executives face constant inquiries from buyers looking to acquire their historical archives.

The Data Moat is the Only Moat Left

When tools and model checkpoints are available to everyone, accumulated data history becomes the surviving defense against commoditization. Ehrlich summarized the reality: “the only real advantage that company have today is of course their people but also the data that they've approved because everyone has access to all of those cool new tools. Yeah, it's become a moat.”

If your software runs workflows, every logged interaction increases in value. Distressed assets that once had zero salvage value outside physical hardware now hold multi-million-dollar training value for foundation model builders.

What to Do With This

Audit your application's database logs by Friday. Stop purging historical user interactions, workflow corrections, and multi-step transaction logs to save on cloud storage costs. Separate your execution telemetry from temporary system logs so you preserve your domain-specific business data.