“In the next two to four years, this is going to be over a trillion dollars a year,” says Ansari, chief executive officer of data acquisition firm Micro1. The financial projections from infrastructure brokers reflect a fundamental shift in how frontier models are constructed. With public data exhausted, artificial intelligence laboratories are spending over $45 billion on private operational files to train future architectures.
Brokers and model developers are now approaching operating businesses with direct acquisition offers. According to broker proposals sent to media organizations like Wired, developers of Anthropic and OpenAI's caliber are allegedly offering between $350,000 and $1.8 million as a one-time payment to acquire historical internal archives, including documents, emails, Slack messages, and support tickets. These proprietary archives capture internal operational routines that synthetic benchmarks and web scrapes cannot replicate, turning daily enterprise communication into an asset class.
The Commercial Rush for Enterprise Workflows
The supply side of the market is expanding as specialized brokers position themselves between corporate repositories and frontier laboratories. Industry players Handshake, Micro1, and Mercor are actively acquiring private databases across various sectors. Mercor, established in 2023, saw its three cofounders become billionaires at 22 following the demand for corporate data assets. The firm originally operated as an AI recruitment outfit before pivoting when its management recognized that major laboratories were dissatisfied with training data quality.
Mercor chief executive officer Brendan Foody stated that the market is poised to explode as the AI industry demands dramatically higher data volumes. Handshake has taken an equivalent commercial route by licensing professional documents across various disciplines, offering individual contributors payments such as $1,000 for story materials and building a major revenue stream by purchasing graduate students' coursework and papers.
"We license professional documents across a lot of fields so the labs can test and train how models handle real knowledge work," a Handshake representative explained, confirming that academic papers and editorial files are routinely packaged for commercial laboratories.
This demand has rapidly reshaped startup formation. Venture activity has followed these capital expenditures: in the most recent YC batch, 40 of the 65 launched startups are classified as data companies. While Micro1 deploys acquired information internally to build proprietary models, it also resells access to external AI firms.
How far down the corporate solvency spectrum does this aggressive purchasing drive reach?
Insolvency Sales and Anonymization Limits
Bankruptcy proceedings have become a primary liquidation venue for internal records. When Spirit Airlines entered bankruptcy, the carrier placed its entire digital estate up for sale. The asset package comprised 80,000 corporate email accounts, 500 million internal messages, 20.6 million Microsoft SharePoint documents, and 1 million time-card records. Google submitted an initial bid of $10 million for the corpus, but Micro1 surpassed that offer by $2.5 million to secure the inventory.
Data brokers maintain that rigorous sanitization protects former employees and partners from exposure. Intermediaries pitching liquidation services to failing startups claim they strip out all personal identifying details before selling archives to foundation model builders. Mercor also claims that inbound data streams undergo auditable de-identification protocols to neutralize corporate compliance concerns.
These automated scrubbing assurances face direct legal challenges in court. Meanwhile, Anthropic acknowledged the substantial commercial market for data while stating that it does not acquire workflow materials from operating organizations such as Wired.
Audit internal document retention policies and employee communication platforms to verify whether corporate wind-down agreements or software licenses permit the liquidation of active message archives to third-party data aggregators.
Against the field
Why are foundation model developers paying millions for ordinary corporate chats and internal support tickets? Public web data and synthetic text hit a quality wall, turning messy workplace records into high-value training material through middlemen like Micro1 and Mercor. Yet an acute tension sits right inside this trade: while intermediaries promise verifiable scrubbing of personal information, their de-identification protocols are already facing challenges in court.