AI Companies Really, Really Want to Buy Your Work Files and Emails
AI-assisted digest, checked against the original source before publishing. How we work
Data brokers are quietly buying internal corporate files, emails, and coursework to train AI models — an acquisition boom some industry leaders say could rival spending on compute.
Data brokers have been pitching companies directly for their internal work files, emails, and call transcripts, with one offer running as high as $2 million. The pitches cite an industry where, in one broker's words, "public data is exhausted" and labs are spending tens of billions on private data to build better models.
The firms driving the boom are only a few years old. Handshake, Micro1, and Mercor are described as the leaders, with Mercor — founded in 2023 — the apparent front-runner; its three cofounders became billionaires last year at age 22. Handshake has offered a freelance writer $1,000 for a single story and confirmed that it licenses professional documents so labs can train models on real knowledge work, including journalism and graduate students' papers and coursework. Other startups, like a company called Sunset, pitch themselves as winding down defunct startups and brokering scrubbed, anonymized data to labs.
Executives frame the market as boundless. Mercor CEO Brendan Foody argues Big AI has two bottomless needs, compute and data, and that hiring experts to simulate workflow failed to capture the messiness of a real operating business. Micro1 CEO Ali Ansari says labs, enterprises, and robotics firms currently pay roughly $40 to $50 billion a year for data — a figure the reporter could not verify — and predicts the market will top a trillion dollars a year within two to four years. He also claims about half of Y Combinator startups are data companies; in fact, 40 of the most recent batch's 65 companies are classified as data startups.
The practice's scale was exposed when bankrupt Spirit Airlines auctioned digital assets: 80,000 email accounts, 500 million internal messages, 20.6 million SharePoint files, and a million time-card records. Google bid $10 million, then Micro1 topped it by $2.5 million, sparking a dispute over the winner; the sale is now on hold after the flight attendants' union objected that their private information was included. EPIC general counsel Mayu Tobin-Miyaji argues anonymization may not eliminate risk, since unstructured emails and Teams messages can contain references to medical appointments or family matters.
Venture capitalist Gordon Ritter contends the industry has not set norms or standards for data acquisition and is moving faster than society's ability to establish them. What to watch: whether the Spirit sale and similar disputes produce enforceable rules, and whether labs' stated de-identification claims hold up as more live-company workflows hit the market.
keyFacts:[
Mercor founded 2023; three cofounders became billionaires last year at 22
Micro1 CEO says labs pay $40–50B/year for data; unverified
Key facts
- • Another email with a similar pitch (adding call transcripts to the list of desired information) raised the stakes to.
- • “Public data is exhausted, and the labs are spending over $45B on private data to build better models,” writes.
- • A freelance writer recently wrote about her experience with a company called Handshake, which offered her $1,000 for.
- • Industry leaders Handshake, Micro1, and Mercor are snapping up databases as if it were a modern gold rush.
Why it matters
Builders who assumed internal documents were safely out of AI vendors' reach should audit data-provenance terms in vendor contracts and wind-down deals. Expect procurement, legal, and works-council scrutiny to tighten — the Spirit Airlines union challenge shows employees can contest sales of their data retroactively. For AI teams, the pipeline offers rich training signal but rising regulatory and
Spotted an error? Send a correction.
Get the day's AI news in one email
The Daily Epoch · 60-second read · unsubscribe anytime
