One man’s digital trash is an LLM’s treasure.
There is a gold rush in the growing business of AI data brokers. People are purchasing data — ranging from a defunct small business’s Slack messages or employee presentations to years of shop-floor records from former manufacturers — and then selling or licensing it to frontier AI labs to train their large language models.
These data brokers aren’t rag-and-bone men. They’re companies like Micro1 and Mercor that have launched in just the last two years and are already valued in the billions.
And they’re just getting started.
Micro1 CEO Ali Ansari estimates that annual spending on data by AI labs, enterprises and robotics companies has grown from roughly $2 billion a few years ago to nearly $50 billion today — and he predicts that spending could approach $1 trillion annually within the next few years.
“We believe this market is going to be one of the largest markets in the world. It’s also going to be one of the largest job sectors,” Ansari told The Post.
The industry is one of a host of new businesses springing up around AI. Brokers typically take a cut of sales — and successful ones are bragging that they can make millions in a single week.
“It can be quite lucrative,” Ansari said, adding that he has heard the practice has helped some businesses avoid bankruptcy. Companies are also reaching out to him in hopes of getting paid for their data.
Around 200 businesses a day sign up to explore licensing their data to Micro1, though the smallest ones rarely make it through the process because they simply don’t produce data with enough volume or complexity to make it worth buying.
Most businesses Micro1 works with have between 30 and 150 employees, Ansari said, and payments can range from a couple hundred thousand dollars to a few million, depending on the amount and value of the data. That can potentially represent a substantial share of a small company’s annual revenue.
According to Ansari, the most valuable data is typically from newer companies with the largest volume and range of information — numerous customers, employees and operating histories.
Even businesses that have already gone belly up are getting in on the deals. Spirit Airlines — which filed for Chapter 11 in late 2024 and went out of business earlier this year — was one of the notable companies to sell its operational data, reportedly for around $10 million. The company’s data was one of its most valuable remnants.
Ansari believes that the nascent industry will soon become a standard way every company makes money. “I think we’re going to see a similar thing where data licensing becomes as common as advertising for businesses to just have like a revenue line item that touches on data licensing,” he said. “We’re just in the very early days of that.”
The data itself, sources told me, is anonymized. Names and identifying details are replaced with generic information so as not to give away personal data or be traced back to a specific person — or at least that is the hope. The purpose of the information is to train agents, not keep track of individuals.
“Of course, the caveat is that these systems are not perfect,” Ansari admitted.
Demand for the date is expected to be ongoing. Even as the models improve, new knowledge and specialized knowledge — updated laws, medical discoveries, new business tools — will always need to be shared with frontier labs. In fact, an LLM’s knowledge of those advancements is increasingly what can differentiate one AI tool from another.
Ansari is confident the need for data will only grow as models improve. He said, “It’s more so like an infinite last mile.”