Leadership2 publishers3 min readPublished
Google outbid Mercor by $2.5 million for a dead airline's Teams messages, calendars and codebase. Leaders need a data-disposition position before the wind-down, not during it.
The Board Room · Leadership desk

Compiled by The Board RoomSomething wrong?How this is made
Google agreed to pay about $10 million for Spirit Airlines' internal corporate data, according to a bankruptcy court filing dated August 14 [1]. Spirit stopped flying in May after going bankrupt under high jet fuel prices and steep debts [2][3]; the email outlived the airline and cleared a competitive auction.
The winning bid beat data-labelling startup Mercor's $7.5 million [4], a premium of roughly 33 percent [1]. Per the filing, Google gets internal documents, workflows, emails, some 500 million Microsoft Teams messages and Spirit's codebase [5]. Forbes, citing News Nation, put the email count at roughly 10 million alongside the 500 million Teams chats, plus spreadsheets and calendars [6]; a separate Forbes account described "hundreds of millions" of employee emails and operational data [7]. The counts vary by report; the category does not.
Customer records were carved out. Nearly 100 million passenger profiles and about 50 million Free Spirit loyalty records are excluded, and Google has said the set excludes customer data, is anonymized, and will be scrubbed of personal information [8][9].
Spirit reported about 9,700 full-time employees in 2025 [10]. On those figures the workplace record of an entire company sold for roughly $1,030 per employee [2], or about 51,500 Teams messages per head [3]. That is the number to put in front of your own board, because the buy side is broadening. Labs had already consumed the public internet by late 2024 and now want operational data [11]. Mercor's own site is blunt about it: "Frontier AI labs can't build enterprise-grade models from public data alone," and the data on how teams communicate, close deals and ship software "lives inside your company" [12].
Wind-down services SimpleClosure and Sunset now offer to buy client data for AI training [13]. Micro1 has offered 50 midsize companies between $100,000 and $2 million for access to anonymized data [14]. Handshake AI is going straight to individuals with up to $30,000 for high-quality written documents, stipulating the seller must own them and be authorized to share [15]. Jared Pope, an employment attorney and CEO of Work Shield, told Business Insider he has spoken with CEOs already fielding calls from would-be buyers, and that it is generally the company's decision whether to sell de-identified data [16][17].
Against that, the internal apparatus is thin. George Socha, an attorney at e-discovery firm Reveal, told Business Insider there is no single standard place where companies set out their policies on employee data and workplace surveillance, and that firms ought to establish reasonable boundaries and convey them clearly to staff [18][19]. The commercial logic favours getting this right early: $10 million is modest next to the litigation Anthropic has faced over pirated book data, some of it still running [20]. Clean title is most of what Google bought.
Two boundaries worth knowing. Vanessa Matsis-McCready of Engage PEO says sensitive personal information such as Social Security numbers is likely protected by state laws even if the employer fails, and that bankruptcy law will not override that [21]. Ordinary work product is another matter. As Pope put it: "Assume anything you create on a company system could outlive your own job there" [22].
What to watch: whether data starts appearing as a named line item in bankruptcy schedules and asset purchase agreements rather than a residual; what your retention schedule and vendor renewals actually permit a counterparty to resell; and how staff react when the policy is stated out loud. Elon Musk told SpaceX workers their data will be used to train Grok, saying it "will inherit your thoughts and ideas" [23]. Whoever makes that announcement owns the consequence.
Ranked by verification strength, evidence, and original report placement.
Google won a bidding war, paying $10 million for Spirit Airlines' internal data, according to a bankruptcy court filing dated August 14.
Certain components are excluded from the exchange: nearly 100 million passenger profiles and some 50 million records related to Spirit's Free Spirit loyalty program; Google is agreeing to scrub records for personal information.
The dataset excludes customer data and has been anonymized, Google said.
Enterprise data has become useful for AI training because labs had already hoovered up the entire public internet by late 2024.
Mercor's website states: "Frontier AI labs can't build enterprise-grade models from public data alone. The next leap in AI capability requires operational data: how teams communicate, close deals, ship software, and run finance. That data lives inside your company."
Google won out over data labelling startup Mercor, which offered $7.5 million for the Spirit dataset.
Follow any of these and your For You feed starts watching them — no settings page required.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Court-filing backbone, contested dataset details
The core transaction is unusually well grounded for a data-market story: a dated Chapter 11 filing, a named price, a documented bid ladder, an on-record Google spokesperson, and four independent write-ups across two publishers that agree on the $10M/$7.5M outcome. Evidence weakens on the specifics that matter most for interpretation — email volume differs by an order of magnitude between accounts, the anonymization standard is asserted by the buyer rather than verified, and the surrounding market signals are single-sourced program announcements.
One closed deal plus an early buyer market
This is a completed, court-documented transaction with a second serious bidder — stronger than an announcement — and it sits alongside several disclosed programs on both the buy side (Micro1, Handshake AI, wind-down resellers) and the internal side (SpaceX/Grok). But adoption stops there: no evidence any model was actually trained on the Spirit corpus, no count of solvent enterprises that have sold, and Meta's suspended keystroke program shows the practice can stall on employee pushback.
Modestly overstated generalization from one distressed sale
The facts are real and the direction is right, but the framing runs ahead of them. One bankruptcy estate with no going-concern reputational cost sold anonymized internal records; that is generalized into internal comms being a balance-sheet asset for companies broadly, and into a quantified per-employee value that is really the output of a single two-bidder auction. Countervailing detail in the same sourcing — customer carve-outs, state-law protection for sensitive PII, a suspended Meta program — gets less weight than the headline number, and the 'cheaper than litigation' economics are asserted without figures.
Vendor and advisor incentives run through the sourcing
Nearly every non-filing voice has a commercial position in this story: Mercor's own marketing copy is quoted as evidence that operational data is the next capability unlock; the buyer characterizes its own anonymization; and the expert commentary comes from an HR misconduct platform CEO, an e-discovery vendor SVP and a PEO's associate general counsel — all sellers of workplace-data governance services. Forbes' newsletter places the deal directly among valuations and funding rounds. This does not make the claims wrong, but the interpretive layer is supplied by parties who benefit from the trend being large.
High on the transaction, moderate on the trend
Confidence is high that Google bought Spirit's internal corporate data for $10M over Mercor's $7.5M and that a buyer market for work data is forming, because those rest on court filings, an on-record buyer statement and multiple independent items. Confidence is moderate on scale, safeguards and generalizability: dataset volumes conflict, anonymization is unverified, use and resale terms are undisclosed, and the wider trend is evidenced by single-source program announcements from interested parties.
product
Google paid $10M for a dead airline's inbox. Your retention policy is now an estate plan4 publishers
build
A bankrupt airline just set the public price for "how we actually worked": $10M, now $12.5M3 publishers
leadership
Google bid $10m for a dead airline's inbox. That is now the comp for your internal records1 publisher
invest
Google's $10M for Spirit's inbox and code gives bankrupt firms' data a market price1 publisher
Publishers with included, body-backed reporting in this cluster.
3 articles · August 19, 2026
2 articles · August 19, 2026