Court delays Google's purchase of Spirit Airlines data after union objection
A functioning secondary market for data from shut-down start-ups is taking shape: broker platforms are selling code, email archives and customer records from dead companies to AI labs. But a ruling in the Spirit Airlines bankruptcy shows that buying a database is not the same as buying the right to use it — and the legal questions are now having concrete consequences for the transactions themselves.
The undertakers get an industry
When founder Keelan Miskell's crypto start-up died, the wind-down was handled by SimpleClosure — a company that is part of a new growth sector in tech: businesses serving dying start-ups. According to MIT Technology Review, Miskell received around $25,000 for his code base, with a still-outstanding sum — likely around $15,000 — for the workspace data held in Google, Notion, Figma and Slack (MIT Technology Review).
SimpleClosure launched data sales as a service in April. The company's platform, Asset Hub, helps founders sell code and workspace data — emails, chats, documents. According to the company and CEO Dori Yona, the platform has handled more than 150 deals, and 47 percent of them now include the sale of code or data. The company expects to have returned over $10 million to start-ups by the end of 2026 (same source).
The supply is growing too. Estimates from Carta and AngelList show that the number of start-up shutdowns has increased by between 26 and 56 percent annually in recent years. The raw material of this market is thus getting larger with every week that passes.
The Spirit Airlines price tag
The clearest single event so far is Spirit Airlines. In August, as part of the airline's bankruptcy, Google outbid the AI data company Mercor and bought the company's data archives for $10 million. According to MIT Technology Review, the purchase includes 100 million emails, 500 million Teams messages and 176,000 employee records.
Here, however, the sources disagree on scope. Gizmodo describes the archive as "hundreds of thousands of employee records" — a vaguer and possibly larger figure — plus additional categories MIT Technology Review does not mention: one million time-card records and 17 million individually owned OneDrive items, including tax returns, employment contracts and litigation documents, as well as data on flight operations and employee productivity (Gizmodo). The discrepancy is unresolved; what is certain is that the archive is both larger and more sensitive than "emails and chats" suggests.
The court does not simply say yes
The purchase has not been approved. The union Association of Flight Attendants, which represents former Spirit employees, has filed an objection with the US Bankruptcy Court in the Southern District of New York. The argument is that the deidentification standards in the sale protect consumers but leave the employees' private data unprotected (Gizmodo).
The court has postponed approval of Google's purchase. No final outcome exists in the available source material — neither the court's reasoning nor how the case ends is known. But the precedent is already clear: an objection about who the data is actually about can stall a $10 million deal.
SpaceX: informal talks, no deal
The biggest name connected to the market is SpaceX. Bloomberg reported on September 17, 2026, citing anonymous sources familiar with the matter, that the company has held internal discussions about buying customer and operational information from struggling or defunct start-ups — as a cheaper data source for improving its AI models. The talks are informal, could end without a deal, and SpaceX did not respond to Bloomberg's request for comment (Bloomberg via Yahoo Finance).
According to Bloomberg, Elon Musk told personnel at a recent internal meeting that Grok will be trained on "the sum total of all SpaceX information" — adding: "It'll be trained on you."
Bloomberg also reports that SpaceX's AI team has undergone a leadership change: Jack Garabedian, with several years of background at Starlink, has taken over from engineer Diego Pasini, and has since established processes and clear goals for AI training data. The company still uses so-called AI tutors for specialized skills such as finance, science and humor, but the team has experienced departures and leadership changes (Crypto Briefing). Gizmodo furthermore characterizes SpaceX's first quarterly report as saying the company has pledged $15.8 billion to its AI push, while the AI division ended the quarter with $1.3 billion in net operating losses — figures that here rest on Gizmodo's rendering of the report, not on the document itself (Gizmodo).
The database does not come with consent
The analytical core of the story is that ownership of data and the right to use it are two different things. In an analysis by The Next Web, which builds on the European Data Protection Board's (EDPB) 2024 opinion on legitimate interest, the point is made that consent does not travel with the database: the buyer becomes a new controller and must find its own legal basis for the processing — "a harder problem than the price" (The Next Web).
It is important to be precise here: this is analysis, not a regulatory ruling. No data protection authority has yet specifically addressed the sale of data from defunct start-ups. But regulatory attention toward at least one of the players already exists — Ireland's Data Protection Commission opened an inquiry in April 2025 into whether public posts from users in the EU/EEA were lawfully processed to train the Grok models (same source). For European customer databases, which often include people who never consented to anything beyond a single contract with a now-dead company, it remains unresolved whether any buyer could ever lawfully use them for model training.
What no one has proven
Then there is the most fundamental question: do these data even work? According to MIT Technology Review, the extent of performance gains is unclear — no company has so far published tests comparing training results with data from real, defunct companies against alternatives. Nick Heiner of Surge AI, itself a buyer in this market, points out that prices vary wildly because valuations rest on "research theses" — that is, assumptions about what the data might contribute, not documented gains (MIT Technology Review).
In other words: a market with measurable prices — $25,000 for a code base, $15,000 for a workspace, $10 million for an airline's archives — is pricing a benefit no one has publicly demonstrated. Perhaps it is reasonable speculation in an industry where high-quality data is a bottleneck. Perhaps the court fight over Spirit is a reminder that part of the goods — employees' privacy — was never for sale. Neither question has an obvious answer yet. What is certain is that the market has found a name, an infrastructure and a price list — and that the legal questions now come with the bidding.