The Data Wars: AI Data Broker Says Deals Now Run $1.1M-$1.3M

Erik Newsham, founder of Grepped, a startup that buys private company data and sells it to AI labs, wrote on X on 24 October 2026 that the company’s average deal used to be “around $250-$400k” and is “nowadays … closer to $1.1-$1.3m.” Those are the founder’s own figures. No buyer, seller or contract is named, and we have not verified them.

We read the post as one data point in a wider fight over AI training data, which our October reporting has covered from several sides: who pays for data, who sues over it, who fences it off and who opens a market for it. This piece sets them side by side. The figures come from different sources and measure different things, so they are not combined.

This Week In The AI Supercycle · VIDEO

The Business Engineer’s explainer for the week of 18 to 24 October 2026: the companies selling AI’s parts say they are booked for years, while the companies paying for the build found money dearer. It was made before Newsham’s post and does not discuss it.

What the Founder Says

Newsham says Grepped has been “targeting higher end companies willing to sell their data.” According to the post, a referral partner the post calls John, whose father owns a manufacturing business in Indiana, had the father ask fellow manufacturers whether they wanted to license their data, and “around 8 have said yes. 4 have closed so far.”

Newsham says most of them were “decently small (60-130ish employees)” and that Grepped “paid around $500,000-$1,200,000 for a 2 year data license.” The post says the referral partner earns a cut of every deal, “six figures each transaction,” and that Grepped has agreed “a 3-year, exclusive license” with a much larger manufacturer that has not yet closed.

The post does not name the manufacturers, the AI labs that buy the data, or the terms of any contract.

What Was Posted on X

Erik Newsham, founder of Grepped, wrote on X on 24 October 2026 that the company’s average data deal used to be “around $250-$400k” and is now “closer to $1.1-$1.3m.” The post describes data licenses with manufacturers brought in by a referral partner it calls John. The figures are the founder’s own and unverified.

Brendan Foody, chief executive of Mercor, wrote on X on 23 October 2026 that Mercor will pay “up to $100K for every company you refer that licenses its data” and that it has “partnered with > 500 companies so far.” These are the company’s own figures.

The account @GeniusGTX on X re-shared, on 24 October 2026, remarks Foody made on stage at TechCrunch Disrupt 2025. We quote those remarks from TechCrunch’s report of 29 October 2025.

What Grepped Says It Does

Grepped’s website, grepped.ai, offers “premium licensing for company data worth millions.” It says today’s best AI systems “finish under 5% of the actual real jobs experts put in front of them” and need “real examples of how people take a task from start to finish.” The site lists messages, codebases, documents and raw records as data types, and says Grepped strips “customer names, contact details, and identifying information before anything leaves processing.”

ARR Club reported on 6 October that Newsham announced Grepped hit “$4M MRR ($48M ARR run rate)” two weeks after entering the data business, and that the company sells to AI labs. That figure is also self-reported.

The Other Bidder: Mercor

Grepped is not the only company bidding for company data. Brendan Foody, chief executive of Mercor, wrote on X on 23 October 2026: “We’ll pay you up to $100K for every company you refer that licenses its data to @Mercor.” The post adds: “We’ve partnered with > 500 companies so far,” and asks for companies “with 100+ employees and data in code, gaming, manufacturing, healthcare or professional services.”

Mercor’s referral page, which we read on 25 October, says Mercor has paid “$50M+” to companies for their data and “$2M+” to referrers in 2026. It lists 24 kinds of company. For a manufacturing and industrial company licensing “production, maintenance and inspection records from the shop floor,” the page says the company earns $100k–$1.5M and the referrer up to $85,000. These are Mercor’s own figures.

The same company also pays people. A post on X on 24 October 2026 by the account @GeniusGTX re-shared remarks Foody made on stage at TechCrunch Disrupt 2025, which TechCrunch reported on 29 October 2025. According to TechCrunch, Foody said: “There is an argument that Goldman Sachs doesn’t love the idea of having models that are able to automate their value chain.” He added: “Their customers don’t want to give them data to automate large portions of their value chains, so they need to hire contractors who previously worked at those companies, understand those workflows, and are willing to train models to automate them.”

TechCrunch reported then that Mercor pays industry experts up to $200 an hour, per Foody, and had raised funding at a $10 billion valuation. In July 2026, TechCrunch reported, citing Bloomberg, that Mercor was in talks to raise at a $20 billion valuation. Our expert-pay piece reported Bloomberg’s figure of $110 an hour as Mercor’s average pay.

The Data Wars, Side by Side

Grepped’s claims sit next to Mercor’s and to four other fronts our recent pieces have reported.

Sources: as listed in each row. The rows measure different things (prices, legal claims, terms, forecasts) and are set side by side, not combined.

Who Sues Over Data

Reservoir filed a complaint against Anthropic on 21 October over 853 songs and asked for statutory damages that add up to about $151.4 million, by our sum of the three amounts in its complaint, as our music lawsuits piece reported. Rolling Stone reported that the Elvis Presley suit “is seeking $150,000 per infringed track, or roughly $41 million.” Both are claims, and no court has ruled on them.

Who Fences It, and Who Sells It

The Information said on X that Salesforce customers using Claude and other external AI tools “could soon face new charges,” and that some businesses “may move data out of Salesforce if the fees are too high,” as our open-versus-fences piece reported. Meta removed “train” from both English editions of its crawl clause before its terms take effect, as our Meta terms piece reported.

On the other side, DataVendor, a marketplace where companies list private data for sale to AI labs, opened a public beta on 22 October, as our DataVendor piece reported. Its post says: “Data is the fundamental bottleneck for AI progress.” That is the company’s claim.

For scale, Gartner research reported by Techstrong puts 2026 AI spending at $2.67 trillion, with generative AI models at $28.3 billion, about 1% of the total, as our Gartner piece reported.

Connecting the Coverage

Our music lawsuits piece of 24 October reported the Elvis and Reservoir claims against Anthropic. Our open-versus-fences piece set out the government and vendor fences on AI, including the Salesforce update. Our Meta terms piece followed the word “train” leaving Meta’s crawl clause. Our DataVendor piece covered a marketplace for selling private data to AI labs, and our Gartner piece put AI spending on models at about 1% of the total. Our expert-pay piece reported Bloomberg’s account of specialists paid to train AI. What this piece adds is a seller-side price: the deal sizes Grepped’s founder says the company now pays, and the referral fees Mercor offers.

The Structural Read

As we read it, one front is price: Newsham says the average deal moved from “around $250-$400k” to “closer to $1.1-$1.3m,” and Grepped’s website says AI systems need “real examples of how people take a task from start to finish.”

As we read it, another front is legal: Reservoir asks for about $151.4 million by our sum, and Rolling Stone puts the Elvis suit at roughly $41 million. Both are claims, not rulings.

As we read it, a third front is access: The Information said Salesforce customers using outside AI tools “could soon face new charges,” while DataVendor opened a public beta for selling private data to AI labs.

Three Implications

FOR COMPANIES HOLDING DATA Newsham says most of the manufacturers that licensed data to Grepped were “decently small (60-130ish employees).”

FOR AI LABS Reservoir’s complaint over 853 songs asks for about $151.4 million by our sum; no court has ruled.

FOR PLATFORMS The Information said some Salesforce customers “may move data out of Salesforce if the fees are too high.”

The Business Engineer Lens

This story maps onto the Business Engineer framework The Four Intelligence Moats.

The framework puts it this way: “The Corpus Moat — built by pretraining, eroding as the public web is exhausted and synthetic data closes gaps.” It adds: “The Container Moat — built by closed data loops inside a customer’s environment, nascent but the deepest position in the stack.”

As we read it, Grepped’s pitch sits between those two lines: its website describes public data as already scraped and trained on, and offers to license the private records companies hold. The framework was published in July 2026 and does not discuss Grepped or these figures.

What Is Not Established

Every Grepped figure here comes from its founder’s posts and website; none has been confirmed by a buyer, a seller or a filing. The post names no manufacturer, AI lab or contract. ARR Club’s run-rate figure is also self-reported. Mercor’s partner count, payout totals and listing ranges are the company’s own. Foody’s Disrupt remarks date from October 2025. The lawsuit amounts are claims, not rulings. We did not contact Grepped, Newsham, Mercor, Foody or the other companies named.

The Bottom Line

Grepped’s founder says its average data deal has moved from “around $250-$400k” to “closer to $1.1-$1.3m,” and that for the manufacturer deals Grepped “paid around $500,000-$1,200,000 for a 2 year data license.” In the same weeks, our reporting recorded lawsuits claiming damages over training data, vendors and platforms changing terms, and a new marketplace for selling data to AI labs.