Data Brokers Are Allegedly Making $2M a Week. Here Is What Is Documented

The two posts#

On October 14, 2026, Greg Isenberg posted a list of observations from 48 hours in San Francisco with data labs and AI founders. The first one reads, in part, "Sooo MANY stories of data brokers making like $2M a WEEK selling company data to labs. I can't tell if it's the beginning or a bubble."¹

Greg Isenberg@gregisenberg

I spent 48 hours in SF hanging out with billionaires, data labs and AI founders and execs running $1B to $100B companies.

My observations:

1. Sooo MANY stories of data brokers making like $2M a WEEK selling company data to labs. I can't tell if it's the beginning or a bubble,…

The next day, levelsio wrote on X and on his blog that data brokers are now buying SaaS datasets for anywhere from $100,000 to $2 million, that struggling SaaS companies like the extra cash, and that the data is resold to AI training companies and frontier labs.²

@levelsio@levelsio

Data brokers are now buying up SaaS data sets for anywhere from $100K to $2M

SaaS are struggling now and some extra money to fill the gap by declining revenues is nice

That data is then re-sold to AI training companies and AI frontier labs for the exact purpose of integrating…

Both posts spread fast. Isenberg's had about 2.5 million views and 21,000 likes when we checked.¹ The levelsio blog post showed more than 225,000 views at the time of capture.² If you run a company with years of support tickets, Slack history or CRM records, it is natural to read them and wonder whether you are sitting on a six or seven figure asset.

This page checks both claims against what is public. We are a broker, so we have an interest in this market, and you should read our numbers with that in mind. Every figure below links to a source, and where sources disagree we say so. We did not find a public record that confirms either post as written, and we did not find one that disproves them either.

What each post actually claims#

The two posts make different claims, and they should not be lumped together.

Isenberg reports a pattern of stories. He does not say he saw books or contracts. He says he heard a lot of stories at events over two days, and he says himself he cannot tell whether it is real or a bubble.¹ The figure is "$2M a week," and he does not say whether that is gross sales, broker revenue, or profit. Those are three very different numbers.

levelsio states a price range for a dataset: $100,000 to $2 million. His post names no broker, buyer, SaaS company or deal, links no report, and does not say how he knows. It reads as commentary, not reporting on a specific transaction.²

Neither is necessarily wrong. A claim with no named source is just unchecked. The useful question is what the public record says about the same numbers.

What is documented, in order of size#

Here is every dollar figure we could tie to a source, from smallest to largest. Treat the table as a map of the evidence, not a price list for your data.

Table 1. Public dollar figures for company data sales and offers, as of October 22, 2026.

A few notes on how to read this.

The lowest rows are people who shut a company down and sold what was left. Miskell sold a code base for roughly $25,000 and expected about $15,000 more for workspace data from Google, Notion, Figma and Slack.³ SimpleClosure, a firm that helps founders wind companies down and sells their code and data, told Forbes it had done about 100 such deals in a year with payouts of $10,000 to $100,000.⁴ Those figures sit below levelsio's floor of $100,000.

The middle rows are offers from one buyer, micro1. Its tiers start above $100,000 and the top tier is above $1 million for highly unique, proprietary data.⁵ Business Insider, as relayed by AI Weekly, reports micro1 pays $100,000 to $2 million and often sends six figure offers that expire in 48 hours.⁶ That is the closest public match to levelsio's range. It is one company's numbers, and we found no public record of a closed micro1 data purchase from an ordinary company. See our comparison of Mercor and micro1 for what is known about each.

The top rows are rare and unusual. Spirit Airlines was a large airline whose archive, per Troveo's explainer, covers about 34 years. It reportedly includes about 100 million emails, 500 million Teams messages and about 30 million lines of code.⁸ Reddit holds years of text written by its users, and the $60 million figure is anonymously sourced.⁹ Neither is a typical company, and neither is a good guide to what your data would fetch.

For the details on each, see the deal pages for Spirit Airlines and Google, Cielo24, Keelan Miskell, SimpleClosure and Reddit and Google. The full list is on the deal tracker.

Where a $2 million week could come from#

Here are the readings of "$2M a week" that we can test.

1. Gross sales by one very active broker#

The only public number that comes near is Grepped. A report in Dealroom says its founder, Erik Newsham, posted on X on October 13, 2026 that he had entered AI data brokerage two weeks earlier and reached a $4 million monthly run rate, about $48 million a year. He called it "the most lucrative space I think has ever existed."¹¹

Do the arithmetic. $4 million a month times 12 is $48 million a year. Divided by 52, that is about $0.92 million a week, a bit under half of $2 million. Dealroom notes the figures are self-reported, that some people on X were skeptical because the business is only weeks old, and that the same post sought staff at pay bands up to $4 million to $9 million a year for a semiconductors VP.¹¹ The company, as described, lets businesses list internal records such as chat logs, codebases, documents and accounting exports, then de-identifies and packages the data for AI labs and pays sellers within seven days of a sale.¹¹ See the Grepped page.

One number, from one founder, weeks into a business, is not a market. It is also the sort of claim that could be the source of a "$2M a week" story in a room of AI founders, though we cannot say it was.

2. The whole market, not one broker#

$2 million a week is about $104 million a year. If that were spread across the whole data brokerage market it would not be surprising, given that micro1 alone has said it will spend $1 billion over 12 months.⁵ The micro1 figure is a spending commitment funded by Citi and Hercules Capital, not a record of sales, and a commitment is a ceiling, not a result.¹² Still, $1 billion over 52 weeks is about $19 million a week of intended buying. So an industry-wide figure above $2 million a week is not absurd. A single broker making that much is a much bigger claim.

3. Confusion with Mercor's payouts#

Mercor, the AI data and expert marketplace, told TechCrunch it facilitates more than $2 million a day in payouts.¹³ Cryptobriefing, after a 60 Minutes segment on Mercor, estimated daily payouts to contractors somewhere between $1.5 million and $4 million, noting this is an estimate and not a number Mercor reports.¹⁴ The Information reported a gross annualized revenue run rate above $2 billion in June 2026.¹⁵

Those are real, large numbers, and they are easy to mix up with "$2 million a week." But that money goes to people who do work: writing, grading and correcting model outputs. It is not what a broker earns for selling a company's Slack history. Mercor does also run a program that pays companies for workflow data, and publishes no payout amounts.¹⁶ We are not saying anyone made this mistake. We are saying the similarity in numbers is a reason to ask which kind of "$2 million" a person means.

4. Gross sales versus what a broker keeps#

Even if a broker moved $2 million of data in a week, we found no public information on what brokers or platforms keep. Mercor says it covers extraction, anonymization and transfer costs with no setup fees or deductions to the seller.¹⁶ micro1 says nothing about its margin. SimpleClosure's page does not say how it is paid or how proceeds are split.¹⁷ Without that, "making $2 million" could mean anything from a few percent of it to all of it.

What the evidence can and cannot support#

We can say these things with sources:

• Buyers are spending real money on company data. micro1 has said it will commit $1 billion over 12 months and published price tiers.⁵¹²
• Prices for individual deals vary by orders of magnitude, from about $10,000 to $10 million.³⁴⁷
• Some sales are getting scrutiny. More than 110 members of Congress objected to the Spirit sale, and the court hearing has moved to November 6, 2026.¹⁸

We cannot say these things:

• That any broker is making $2 million a week. We found no document, filing or named source for it.
• That a typical company can sell a dataset for $100,000 to $2 million. The public reporting on ordinary companies, as opposed to shut-down startups or special cases, is thin. Our main guide notes that we found no public source naming a buyer or price for an ordinary company's raw support tickets or CRM exports.
• That prices will hold. The MIT Technology Review piece notes that no one has published tests comparing real-world data with synthetic alternatives, that valuations rest on speculative "research theses," and that prices may stabilize in the coming year unless an AI bubble bursts first.³

What this means if you hold company data#

Three practical points.

First, check what you own. Customer contracts, privacy promises, employee data and confidentiality clauses decide what you are allowed to sell. micro1's own page tells companies to have the "internal approval processes and permissions" for participation, which puts the burden of checking on the seller, as one analysis of its terms points out.¹⁹ See our risk checklist for what to review.

Second, expect a lower number than the headline. The documented range for companies that are not famous is much lower than the top of any range quoted online. micro1's own illustrative payout estimator showed $50,284 to $92,776, below its lowest published tier.¹⁹ That is one example, not a quote, but it shows how far an estimator can sit from a headline.

Third, get terms in writing. Exclusivity, field of use, deletion, audit rights and indemnity decide what you have actually sold. A price without those is half a contract.

If you want to sort what you hold before approaching anyone, the assessment walks through the data types and rights questions.

Bottom line#

The "$2 million a week" claim is a story, not a record. The "$100,000 to $2 million per dataset" range overlaps with what one buyer, micro1, says it pays, but it is not a market-wide price, and most documented sales of ordinary company data are smaller. The documented top-end is a handful of unusual deals such as Spirit Airlines, which is still waiting on a court.

If the posts turn out to be right, there will be filings, invoices and named sellers soon enough. Until then, treat the numbers as a reason to look at what you hold, not as a forecast of what you will earn.