A year ago, nobody had a reason to build this directory. Now at least eight companies have public pages that say, more or less, “we will pay you for your company’s records.”
They want the boring stuff. Slack threads. Support tickets. Work orders. CRM notes. The record of how a real business gets work done. AI labs have read the public web. What they lack is how real companies run, and that lives inside companies like yours.
This is a directory post, so I built it the way I’d build any listing page. Same fields for every entry. A link to the source. A date checked. No rankings by gut feel.
How I built this directory
Every entry answers the same five questions:
- What it publishes. The pay range on the buyer’s own page, in the buyer’s own words.
- Size threshold. The smallest company it says it wants.
- Deal structure. One payment, a revenue share, or a mix. Exclusive or not.
- Worth noting. The one detail that sets it apart.
- Source. Each buyer’s name links to the page I read.
I checked every page on October 24, 2026. These programs are new and they change often. Read the source page before you rely on any number here.
The AI data buyer directory
Sorted roughly by the highest figure each buyer publishes.
| Buyer | What it publishes | Size threshold | Deal structure | Worth noting |
|---|---|---|---|---|
| Mode | $100K to $5M, labeled "indicative only" | 30+ employees for most. 10+ for accounting, finance and medical. 6+ for law firms. | Buys an agreed copy of your records, removes identifiers, and resells it for AI research and training | Highest published ceiling. US-based teams fit best. Announced a $5B data initiative on October 13, 2026. |
| Endless | Up to $4M | Roughly 30+ employees, several years of history, documents in English. US, UK and Canada first. | A fixed sum, agreed before anything is shared. Non-exclusive. No commission on the seller's side. | A middleman. It pools data from many companies and places it with several labs. |
| Handshake AI | $100K to $4M+ headline | Its estimator starts at 20 employees and 3 years in business | Scope and terms agreed with you before you commit. The page does not spell out how payment is structured. | Best known as a jobs network. Says it connects businesses with AI labs. |
| micro1 | $200K to $2M+ headline. Tiers of $100K to $500K, $500K to $2M, and $2M+. | 30+ full-time employees. English. US first, then other Western markets. | Reviews your data after delivery, then pays in full within 30 days of acceptance | Announced a $1B, 12-month commitment to buy company data on October 9, 2026. |
| Polyshares | Typically $100K to $2M+ | No headcount minimum. "Any company with real operating history." | Polyshares is the buyer. A license to defined data. Net 30 to 60 after approval. | Its case studies list four licenses. All four were perpetual and exclusive. |
| Replay | $10K to $100K (20 to 50 employees). $100K to $1M (50 to 250). $1M+ (250+). | 20+ employees | Upfront cash plus a perpetual revenue share. New data you produce can keep paying. | The only buyer here that ties its ranges to headcount bands. |
| Scale AI | $10K to $1M+, labeled "illustrative" | 40+ employees preferred. Smaller firms reviewed case by case. | An ongoing partnership, not a one-time transfer. Value scales with how often you deliver. | Names field services (HVAC, plumbing, electrical), staffing, marketing agencies, law firms and insurance brokerage among its industries of interest. |
| Prism by Labelbox | Around $50K to $1M+ | Not stated in its FAQ | Upfront licensing fee at signing plus a revenue share. Exclusive by default. | Every license has a kill clause that forces deletion within 30 days for material breach. Offers wind-down packages. |
Turing also shows up on some lists of buyers. I could not confirm its terms on its own site, so it is not in the table.
If you want a second view of the same market, Company Data keeps a ranked list of what each AI data buyer publishes about pay, with more on how each one pays.
How to read this table
A directory is only useful if you know what the fields mean.
Published ranges are not offers
Look at the labels. Mode calls its range “indicative only.” Scale calls its figure “illustrative.” Endless says its estimate is indicative and the final offer depends on your sources, volume, quality and due diligence.
That is honest. It also means no number in this table is a promise. Every buyer reviews your data before it pays.
“Up to” and “typically” are different claims, too. “Up to $4M” is a ceiling. “Typically $100K to $2M+” is where the buyer says most deals land. Do not compare one to the other as if they mean the same thing.
Headcount is a proxy, not the price
Most buyers list a minimum company size. Replay says it plainly: headcount is a proxy. Years in business, your industry, the apps you run on, and how much data sits in each all move the final number.
A 25-person firm with ten years of clean ticket history may interest a buyer more than a 200-person firm with two years of messy records. Size gets you in the door. The record sets the price.
Track records are self-reported
Polyshares publishes the most specific track record. Its case studies list a $450K license with a roofing company, $300K with a general contractor, $275K with a solar company, and $110K with a global marketing agency. Company names stay confidential.
Those are useful data points. They are also the buyer’s own figures. Nobody outside the deal has confirmed them. Treat any buyer’s “here’s what we paid” as marketing until you see an offer of your own.
Exclusivity changes the math
This is the field most people skip. It matters most.
- Non-exclusive means you can license the same records to another buyer. Endless says its license works this way.
- Exclusive means one buyer gets it. Prism says its deals are exclusive by default. All four Polyshares case studies were perpetual, exclusive licenses.
An exclusive deal closes every other door for that data. If a buyer asks for exclusivity, ask what you get for it. A higher price should come with it.
Upfront cash or a revenue share
Endless, micro1 and Polyshares describe a fixed payment. Replay and Prism pay something up front plus a share of later revenue. Scale talks about recurring partnerships.
A fixed payment is certain. A revenue share can pay more over time, or much less. If you take one, ask how the share is calculated and how you will see the numbers.
The ranges are not a ceiling either
Rare data can draw more than any published range. When Spirit Airlines went through bankruptcy, its internal records went to auction. Google won in August 2026 with a $10 million bid. Mercor was the backup at $7.5 million. micro1 later offered $12.5 million. At last report the court had not approved a sale, after a flight attendants’ union raised privacy objections.
Most companies will never see a bidding war. But published numbers describe a typical deal, not the top of the market.
Before you contact any buyer
A directory gets you a shortlist. It does not get you a deal. Three checks first:
- Check your rights. Your privacy policy and customer contracts decide what you can license. The FTC warned in 2024 that quietly changing your terms of service to allow AI training could be unfair or deceptive. Do not rewrite a policy after the fact to make a sale work.
- Make it an owner’s decision. Company records belong to the company. Only an owner or executive should sign a license.
- Get more than one offer. Buyers want different things. The one that values your records most is rarely the first one you call.
AI labs are also not the only market. Brokers, investors and research firms buy data too, for different reasons. If your records do not fit this table, read about the five kinds of buyers for company data, from brokers to hedge funds before you give up on them.
Why this belongs on a directory site
Most of what I write here is about building good directories. A new market with a handful of players, no standard terms and a lot of confusion is exactly when a plain, sourced, same-fields-for-everyone list is worth the most.
I will update this page as buyers change their terms. If one of these pages changes before I do, trust the buyer’s page.
Frequently asked questions
Who buys company data for AI training?
The buyers in this directory are Mode, Endless, Handshake AI, micro1, Polyshares, Replay, Scale AI and Prism by Labelbox. Each licenses operational records like email, chat, tickets and CRM data for AI training.
Which buyer publishes the highest number?
Mode publishes the highest fixed ceiling, $100K to $5M, and labels it indicative. Handshake AI lists $100K to $4M+, with no upper limit. A published ceiling is not an average payment and not an offer.
Is a small company too small to sell its data?
Not always. Replay's published tiers start at 20 employees, Handshake AI's estimator starts at 20, and Polyshares says it sets no headcount minimum. Most of the others start around 30 or 40 employees.
Should I take an exclusive license?
Only if the price reflects it. An exclusive license stops you from licensing the same records to anyone else. Endless says its license is non-exclusive. Prism says its deals are exclusive by default.