You cannot sell a company’s files while it is still using them.
Everything on this page depends on one fact: Spirit Airlines is gone. It flew its last flight on 2 May 2026, brought down by fuel costs it could not absorb and financing talks that failed.116 What was left behind was a debt, a list of creditors, and a very large pile of files.
The shape of the end. Trouble from 2020 onward, operations ceasing on 2 May 2026, roughly $8.1 billion in debt behind the estate and around 17,000 jobs gone.1 Three months later the files went up for sale.2
When a company fails in the United States, it does not simply switch off.
It enters a court process. A judge takes charge, the company’s remaining property is gathered into a single legal pot called the estate, and that property is sold to raise money for the people the company owes. Aircraft. Gates. Spare parts. Brand names.
And, increasingly, files. Because a company’s records are property too — and property, in a bankruptcy, is for selling.
A number means nothing until you put it beside another number.
Ten million dollars sounds like a lot. Here is the same estate selling other kinds of property in the same collapse.
Areas are proportional to the dollar figures. JetBlue paid $58,500,000 for 22 of Spirit’s take-off and landing slots at one New York airport, LaGuardia.1 The whole data archive went for $10,000,000.1
Twenty-two permissions to land a plane at one airport cost nearly six times what the entire institutional memory of a seventeen-thousand-person company did.
Our arithmetic $58.5 million across 22 slots is about $2.66 million per slot. At that rate, 3.8 slots would have bought everything Spirit ever wrote down. Dividing is ours; the two prices are reported.19
Watch the bidding. The third bid is the one that matters.
Two buyers. Google, and a company called Mercor — a marketplace that buys and sells data used to train AI models.17 Step through it.
The bidding sequence as reported.1 The backup bid is the offer that takes over if the winning deal collapses before it closes.
The same buyer offered more money for the version with the names still in it.
Mercor bid $5.2 million for the cleaned data. Then it bid $7 million for the raw data instead. Same archive. Same buyer. Minutes apart.1
Mercor wanted the unscrubbed version so that it could run the anonymisation itself, with its own tools.1 Whatever the motive, the two offers sit side by side in the record, and the gap between them is a number.
This is the most useful thing on the page for understanding what data is worth, because it is the rare case where a market prices the privacy layer directly. Removing the names makes the archive less valuable to at least one bidder — and here is roughly how much less.
Our arithmetic $7,000,000 minus $5,200,000 is $1,800,000, or about 35% more for the unscrubbed version. The subtraction is ours; both bids are reported.19
What does “an airline’s data” actually mean?
It means this. Every quantity below is a count of individual things — one email, one message, one fuel receipt.1
One bar, drawn to scale, split by what the items are. Click a slice to zoom into it and re-draw the bar using only that slice. Keep going, and watch how far down you have to travel before you reach anything about a person.
And then there is the finance system. 7,510,221,520 revenue transaction records reaching back to May 2008 — every fare, fee and refund the airline ever booked.1
Divide the price by the pile.
$10,000,000 spread across roughly 647,000,000 items. This is what a record is worth when somebody actually has to name a figure.1
Move the sliders. The price per record is fixed at Spirit’s rate; only the size of the paper trail changes. The result is what your own working life would fetch at the same price.19
Nobody at Spirit received that money. Not a share of it, not a notification that the calculation was happening.
Our arithmetic $10,000,000 across 17,000 people is about $588 a head — and all of it went to creditors.19
Nobody can quite agree on what changed hands.
Four large categories are described by some outlets as part of what Google bought and by others as carved out of it — plus a fifth that only ever appears on the excluded side. This page does not pick. It shows you both readings, and who says which.
Switch between the two published readings. The totals at the top move with your choice — which is the point: the size of the thing being sold is not a settled fact in public reporting.
Notice what the two sides agree on. Both readings describe the same categories with the same counts.11 They differ on a single column of the sale schedule: whether the row says included or does not. That is the entire disagreement, and it covers close to seventy million records of customers dealing with a company.
Which is worth sitting with. The most ordinary question a reader can ask — what, exactly, did Google buy? — does not yet have one answer in public.
Passenger records were not in this sale. Read that sentence again slowly.
A very large amount of customer data was kept out of the lot Google bought.110 That is real, and it matters. It is also not the same thing as protection.
Everything here stayed out of Google’s lot. Open the fine print when you get to it.1
“Not in this sale” and “not for sale” are different sentences.
The passenger data was not withheld from the market. It was withheld from this lot. It sits on the shelf, still owned by an estate whose entire remaining purpose is to convert property into money, with a documented buyer appetite and a demonstrated price. Nothing in the arrangement prevents a second sale. The arrangement is what makes a second sale straightforward.
If you flew Spirit, this is the part of the story that is about you. Not “your data was protected.” Rather: your data was not in the box that opened in August.
Every record gets cleaned before Google sees it. Follow who decides what clean means.
De-identification means stripping out the details that point at a named human being — the name itself, the staff number, the home address — so that what is left describes behaviour without identifying the person behaving. The deal requires it, by a third party, before delivery.1 Here is the rest of the arrangement.
The chain of custody exactly as the filing describes it. Select any question to see which party answers it.112
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.
Take that statement at face value. It is probably accurate. The interesting question is not whether Google is telling the truth about receiving no names — it is what the archive still is once the names are gone.
Which brings us to the one requirement that changes everything.
The names come out. The threads stay in.
The deal requires that the links between records survive the cleaning, so the same anonymous person can still be followed from an email to a support ticket to a payslip.1 That requirement has a name — referential integrity — and it is the reason the archive is worth buying. It is also the reason the anonymity is thinner than it sounds.
One employee’s trail through seven company systems. Strip the names, then start adding facts from outside the data and watch how quickly the anonymous person stops being anonymous. The records and the narrowing here are an illustration — the filing publishes no such numbers — but the structure is exactly what the deal requires.1
“Anonymous” and “anonymous but still joined up across twenty years” are not the same promise.
A pile of disconnected records is genuinely hard to attach to a person. There is no shape to it. One anonymous complaint about a delayed flight in 2019 tells you nothing about anybody.
A trail is different. A trail has a shape, and the shape is unusual. This person was hired in a particular month, based at a particular airport, worked a particular four-day sequence with a particular crew, filed an expense in a particular city, and had one attendance conversation with a manager. Each fact on its own describes thousands of people. Stacked together, they describe about one — and the stacking is precisely what referential integrity guarantees will still work after the names are removed.
This is not a hypothetical objection invented for this page. It is the thing the union went to court about.
The privacy terms were written for passengers. The records are full of workers.
The Association of Flight Attendants-CWA, the union representing Spirit’s cabin crew, filed a formal objection to the sale.65 Its argument is not that privacy was ignored. It is that privacy was drafted around the wrong people.
Named in the protections on the left; named in the sale on the right. The emptiness of the right-hand column is the objection.6
Their sharpest point is a practical one, and it needs no legal training to follow.
A workforce is a known, finite list. Its size is public. Its crew bases are public. The shape of a roster is public. So when an “anonymised” crew pairing is joined to an anonymised payroll record and an anonymised mailbox — and referential integrity guarantees they stay joined — the result is trivially re-linkable to a named individual.6
Sitting inside the sale: disciplinary records, payroll history, decades of internal messages, and the workplace paperwork of seventeen thousand careers.61
The objection worked, at least as far as the calendar. The approval hearing set for 19 August did not approve anything.
We will fight this every way possible.
The wider frame Some coverage puts the issue in one line: hundreds of millions of worker-touching records are moving under a bankruptcy statute written in 1978, for selling factories and machinery — not human behavioural records.15
Why is a dead airline’s paperwork worth more than the open internet?
Because the open internet has already been read. Every AI lab has had the public web, Wikipedia, the forums, the published code. What none of them has much of is process.
The public web mostly preserves conclusions: the article, the announcement, the finished answer. An operating company preserves the chain that produced them — and preserves it in records that stay linked to one another.
Companies are sitting on decades of records that show how real work gets done.
Now put section 10 and section 13 next to each other, because they are the same sentence read twice.
The threads that make this archive dangerous are the threads that make it valuable. A model cannot learn how a decision turns into a consequence unless the decision and the consequence are still connected. Break the links and you are left with 647 million disconnected fragments — safer, and worth considerably less.
That is not an accident in the deal. It is the deal. The buyer requires the connections; the privacy provisions are written to survive them.
The privacy policy you agreed to was Spirit’s. Spirit does not exist.
Nothing exotic happened here. Bankruptcy did what bankruptcy is designed to do, to a kind of property it was not designed with in mind.7
Four steps, none of which require anyone to consult the people described in the records.
A privacy promise is a promise made by a company. It is only ever as durable as the company making it. When Spirit stopped existing, the promise did not transfer to a successor or dissolve the records — it simply stopped having anyone behind it, while the records it described carried on being property.
Our arithmetic The whole sale raises $10,000,000 against roughly $8,100,000,000 of debt. That is 0.12%. Every email, every message, every payslip in the archive was converted into money to cover about one eighth of one percent of what the airline owed.19
There is nothing unusual about Spirit except that it failed first.
Every one of these exists at the company you work for right now. The only variable is whether the company is still trading.
Two conditions produced this sale. Both are ordinary.
The first is that a company kept its records — which every company does, because storage is cheap and deleting things is a project nobody is assigned. The second is that the company failed, which happens constantly and is nobody’s plan.
The novel part is only the last step: that a buyer now exists who will pay real money for the exhaust, because the thing it wants to learn is not in books. Once that buyer exists, every corporate archive quietly becomes an asset with a price, and the moment of sale arrives at the one point when the organisation that made the privacy promises is no longer around to be embarrassed by them.
You do not get a vote in that sale. You do not get a share of it. Under this machinery you do not necessarily even get told it happened.
An airline stopped existing. Seventeen thousand people’s working lives were priced at about a cent and a half each, and a court is still deciding.