On the night of May 1, Spirit Airlines' final flight, number 1833, left Detroit and landed in Dallas just after midnight. The airline had carried more than 50,000 passengers that Friday, and then it stopped flying for good. Spirit was the first major American carrier in a quarter century to go out of business outright: two bankruptcy filings in less than a year, a planned reorganization that collapsed, a proposed $500 million government rescue that fell apart, and a jet fuel price surge, tied to the war in Iran, that reporting at the time said roughly doubled the cost of operating its fleet. About 17,000 employees lost their jobs. The planes were sold off one by one.
Three and a half months later, at a virtual auction run by the airline's estate, the winning bid was for something else entirely. Google was selected to buy Spirit's internal business data for $10 million, outbidding the AI data company Mercor, whose $7.5 million offer stood as the backup. The archive includes roughly 100 million internal emails, about 500 million Microsoft Teams messages, some 30 million lines of software code, payroll records reaching back to 1986, and pricing data drawn from 7.2 billion competing flights. Spirit's own announcement of the shutdown had said the airline was "winding down all operations." The archive is what wound up.
The sale is not final. Judge Sean Lane of the U.S. Bankruptcy Court for the Southern District of New York was to approve it this week; the hearing was moved to September 9 after the union representing Spirit's flight attendants filed an objection. That delay, and the objection behind it, is where the deal stops being a curiosity and becomes a market event.
The estate sold the record of the company, not its customers
The court filings describe the package with the granularity of a corporate archive: 17 million OneDrive files, 20.6 million SharePoint items, 667,563 IT tickets, operational records from 763,000 flights, 5 million crew pairings, 1.2 million fuel slips, and records of 787,452 parts purchases, alongside financial, marketing, human-resources, and revenue-management data. Google has said the data will go toward product development and training its AI models.
What the sale leaves out is as carefully drawn as what it includes. Excluded are Spirit's 97.5 million passenger profiles, the roughly 50 million records of its Free Spirit loyalty program, co-branded credit cardholder data, customer call recordings, contact details, and attorney-client communications. A third-party firm, chosen and paid by Google, will strip personal information before the data changes hands, and the cleaning must be certified against California's privacy law standards, with health-related material handled under federal medical-privacy rules.
Notice the line the auction drew. The airline's 7.5 billion passenger transactions are in the package. The identities of the passengers are not. What Google bought is the pattern of the business, how it priced, how it scheduled, how it failed, without the people who flew it. The customer was the one part of the company that did not change hands.
The complete record only goes on sale after the company fails
Start with why the archive exists at all. A healthy company's records are among its most guarded assets. Its pricing models are trade secrets. Its internal mail is discoverable in litigation. Its failed experiments are reputational risk. No company that still needs its secrets sells them, and every company that lives long enough to be studied has reasons to keep its papers closed.
Spirit no longer needed anything. The estate was obliged to preserve the documents: for the lawyers, for the auditors, for the creditors, for the court. The payroll records survived because the law did not permit their destruction. The archive was kept because the company was dying, and it was sold for the same reason. The preservation duty that bankruptcy law imposes on a failing estate is what turned an airline's memory into a saleable asset.
That produces a structural quirk in the market Google has just entered. The purpose, as reported from the filings, is to give its models realistic material for simulating real-world business operations. The most complete records of real-world business operations that ever reach the market are the records of businesses that no longer operate. Completeness and failure arrive together: a company that survives keeps its record sealed, and a company that dies puts its record up for bid at its own estate sale. The more complete the archive, the more certain its owner is gone.
That is worth sitting with. The corpus Google bought contains the airline's decisions up to the end: the pricing of the most aggressive discounter in American aviation, the internal debates about a model that could not hold, the wind-down itself. The realism on offer is realism with a known ending. The data is not corrupted by that fact, but it is defined by it. The final chapter of the archive is the reason the archive is for sale, and the same ending is now inside the training material.
The same links that make the data useful are the ones the union is worried about
The Association of Flight Attendants-CWA, which represents Spirit's cabin crew, objected to the sale and asked the court to restrict the transfer of flight-attendant employee data and to secure further protections for employees. Its stated worry is precise. The sale agreement requires the firm cleaning the data to preserve the links between records, because an email that refers to an IT ticket that refers to a payroll record is only valuable for training if the connections survive the scrub. The union argues that those same connections are what could allow information about individuals or small groups to be reconstructed even after names and identifiers have been removed.
This is the classic de-identification problem, and it now has a price tag. The estate and Google answer with process: an independent firm, certified cleaning, categorical exclusions, and removal of personal information before transfer. None of that is window dressing, and none of it fully answers the union's question, which is about what the cleaned data can be reassembled into. Both positions deserve their place. The estate is under a duty to maximize what creditors recover, and the buyer's stated terms are materially more restrictive than the norm. The union represents workers whose most personal records, including payroll and scheduling histories, are in the package, and its concern is a technical one with a technical answer that no one has yet given in open court.
Privacy objections are an old feature of bankruptcy data sales; customer lists have changed hands in retail liquidations for years, usually over the objections of state officials. This sale is different in kind. It is the first complete corporate archive, including the employee record of a 17,000-person workforce, to be sold into AI training, and the hearing on September 9 is the first test of what the market's standard protections are worth.
The archive sold for less than the landing slots
The price deserves a moment, mostly because of what it is not. Google paid $10 million for the recorded life of an airline that at its peak flew about 5 percent of the country's flights, operated more than 200 aircraft, and traded for a market value near $5.5 billion. The same estate sold 22 takeoff and landing slots at LaGuardia Airport to JetBlue for $58.5 million, below Spirit's own estimate of the slots' worth. The complete archive of the airline sold for less than the landing rights at a single airport.
The bidding was thin. Google opened at $5 million, Mercor raised to $7.5 million, and Google finished at $10 million. Two bidders for the fullest corporate record ever placed on a market. Thin bidding is not the same as low value; it reflects a market that barely exists. This auction set the first visible price for a complete corporate corpus, and data brokers and AI labs will attend the next one. But supply is structurally limited, because the next complete archive on the market will belong to another company that failed. Only failing companies sell the whole record. The price may rise; the source will not change.
The deal's real product is the precedent
Read the transaction forward. Google receives a training corpus whose completeness is unprecedented and whose subject is gone. The estate receives $10 million against $8.1 billion in debt, a fair price for the last thing it had. And the market receives a template: the exclusions, the scrubbing, the certified cleaning, the preserved links, all of it, plus the objection that questions whether the template holds. The union's questions do not end when the hearing ends, because they are now the market's questions. Every future sale of a dead company's records will be measured against this one, and the origin story will be the same every time: a company that could not keep going, and records it had to keep anyway. Whether a model trained on such records learns how a business works, or only how it ends, is the question the archive itself cannot answer.
Primary sources
- Reuters' Dietrich Knauth for the auction result, the $5 million to $7.5 million to $10 million bid sequence, the contents and exclusions of the data package, and the de-identification plan, and separately for the AFA-CWA objection and the postponement of the approval hearing to September 9.
- Computerworld's Viktor Eriksson, drawing on Bloomberg Law, for the scrubbing arrangements, the privacy-law certification requirement, and the provision preserving links across datasets.
- The Associated Press's coverage of the May 2 shutdown for the final flight, the passenger and employee counts, the fuel-price surge, and the airline's history, and CNN Business for the shutdown timeline and industry context.
- Travel Weekly for JetBlue's $58.5 million purchase of the 22 LaGuardia slots.