/ai-search885 words

Google Paid $10 Million for a Bankrupt Airline's Internal Data. The Market Is Just Starting

Google paid $10 million for Spirit Airlines' 31-year internal data corpus. Data vendors now net an estimated $2 million weekly selling company archives to frontier labs.

Frontier Labs Are Buying Up Internal Company Data to Train AI. The Marketplace is Booming Despite Privacy Concerns
Frontier Labs Are Buying Up Internal Company Data to Train AI. The Marketplace is Booming Despite Privacy Concerns穿着拖鞋一路小跑 / pixabay
  • Google paid $10 million in August for bankrupt Spirit Airlines' internal data spanning 31 years.
  • Micro1 bid $12 million and Mercor bid $7.5 million for the same corpus; both bids were rejected.
  • Observers estimate data vendors now net $2 million a week in revenue.
  • DataMedsAI acquired a laboratory of 30,000 ovarian primary cancer tumors for AI-licensed drug discovery.
  • Google announced a plan to anonymize the Spirit Airlines data on Wednesday, after the purchase.

Google paid $10 million for the internal data of bankrupt Spirit Airlines in August — a corpus spanning the airline's 31 years of operations — and opened a market that now nets some data vendors an estimated $2 million a week. Micro1 bid $12 million for the same corpus and lost. Mercor, an AI data annotation firm, bid $7.5 million and also lost. Those numbers describe a new trade: frontier labs buying the daily exhaust of companies — emails, Slack threads, customer service transcripts, sales pipeline records — to train models that speak corporate vernacular.

The story reached the public eye through a bankruptcy docket. But the buying began earlier and has spread fast. Dan Novaes, CEO of Mode.Inc, told Inc. that over the summer Micro1 approached him to license his company's internal memos and Slack exchanges for resale to a lab like OpenAI or Anthropic. He declined. The deal required raw data, and Mode.Inc could not audit or anonymize it first. "We were like, 'We're interested, but can we audit that before it actually gets licensed?' And the answer was no," Novaes said. "The answer is still no for all other companies effectively, as far as I know." Micro1 says it runs its own de-identification process on enterprise datasets. That claim is unverifiable from outside — buyers and sellers both say the process works, and no independent audit is cited anywhere in the market's public record.

Novaes then started BizBuyData.com, a spinoff data vendor that promises audit rights and verified de-identification. "There's an opportunity here to give businesses a bit more control and also a reassurance of the de-identification process, the actual audit of any sensitive documents," he said.

Why do frontier labs want Slack archives?

Columbia Business School professor Dan Wang explains the value. "What these data give insight into are the kind of patterns of communication, collaboration, and coordination that made a company successful," Wang said. "Broader than that, what it shows in many ways is how to reason across the system." In short, internal archives teach models how work actually flows, not just how it is described.

The demand curve has steepened. "What started out as maybe one or two frontier labs experimenting with this over the summer has snowballed into, basically, every frontier lab, and then the Magnificent Seven and the Fortune 50," Novaes said. OpenAI and Anthropic are, in some cases, bypassing human annotators — whose wages range from novice to PhD-level rates — in favor of digital archives.

Who is selling, and for how much?

Middlemen have multiplied in recent months. Startups such as DataVendor AI and Miro Advisory broker tech company data and also buy archives from HVAC businesses, plumbers, coffee shops and accounting firms. "I've never seen anything like it. It's grown incredibly fast," said Mike Lisovetsky, co-founder of Miro Advisory. Pricing scales with headcount, interaction volume and time: more users in Slack, email and CRM systems over more years means more money. That logic explains the $10-million Spirit premium, which Lisovetsky calls the sector's juggernaut deal. Novaes frames the seller's math bluntly: "Basically, you're putting in no more than 10 hours of work to unlock six or seven figures for your business, at best."

The market extends beyond office chatter. DataMedsAI, a health care technology company focused on patient data and prescription management, recently acquired a laboratory of 30,000 ovarian primary cancer tumors — what CEO Gerald Commissioning calls "the world's biggest." DataMeds plans to target new treatment discovery and then license that data to frontier labs. "The buyer of the tumor data can use their AI to create new drugs," Commissioning said.

What happens to employee privacy?

Here the record thins. No regulator has acted. No audit standard exists. Alix Dunn, founder and CEO of tech policy think tank The Maybe, questions the ethics of distressed sales: "I would be curious how it feels for past employees in companies that failed to convert equity into value for their staff, to have tech communication of their hard work sold for a pittance and with disregard of their privacy." She also asks the harder question about bankruptcy: "How do we feel about that data going to the highest bidder in a bankruptcy proceeding?" Google announced a plan to anonymize the Spirit Airlines data on Wednesday — after the sale, not before it.

Vendors describe de-identification as an AI-driven swap of real names for synthetic ones. "Anywhere where it says Mike or Michael or Mikey, there's a fake identity that's made to replace it — like Bob or Robert — that's then swapped out," Lisovetsky said. That is a technique, not a guarantee. Whether it removes all personally identifiable information depends on process details no vendor has published.

Wang notes the structural gap. "This is a space that defies regulation right now," he said. He expects vendors to keep operating in that gray zone while shielding themselves from legal threat. He also doubts the middlemen will last: "I don't expect there to be some dominant intermediary that is channeling private corporate data from one source to a set of frontier labs."

The market now runs on trust and speed. Watch whether regulators, or an actual audit, catch up to either.

via inc.com (Source)

More from Elena Vasquez

Elena Vasquez

Show full bio

Correspondent covering people, launches and day-to-day practice in Search rankings and algorithm updates, generative engine optimization at SERP Journal.

221 articles