Buyer directory

Who buys company data for AI, and on what terms?

32 companies that could license your records: AI labs, data labs, marketplaces and a shutdown service. Each is read from its own pages, and every dollar figure is the company's own claim unless a news report or court record says otherwise.

Information, not legal adviceHow we are paid

Glim holds a folded map with a dotted route and three pins.

Logos identify each company. None of them endorses Vaultfish.

32 companies that buy or license company data for AI, at a glance. Each name opens its card below, which links to the full profile where there is one.
Company	What they do	Who they take	Deal size they state	How you get paid
AI and data labs
Appen	Licenses workflow records to AI labs	No minimum published	Quoted per partnership	On acceptance, no costs to you
Centific	Licenses operating records to AI labs	No minimum, wind-downs too	Written estimate after first call	On acceptance, quarterly refresh
Collinear AI	Licenses code and app history for agent training	No minimum, 15+ at peak pays more	Range quoted after first call	Per contract after delivery
Defined.ai	Marketplace for licensed datasets	Needs consent paperwork	No range published	Licensing fees per sale
Google (content partnerships)	Invite-only pilot paying for non-public content	By invitation, content providers	No public terms	No public terms
Halluminate	Data lab, partnership interest form only	Not published	No terms published	Not published
Handshake AI	Licenses company data to AI labs	20+ full-time staff, 3+ years	$100K to $4M+	License fee, 30 to 60 days after acceptance
Labelbox Prism	Licenses operating data to AI labs	No minimum, wind-downs welcome	About $50K to $1M+	Upfront fee plus revenue share
Mercor	Licenses workflow data it calls anonymized to AI buyers	No minimum stated	Up to $300K reported	License fee, no fees to you
micro1	Company-data partnerships for AI development	30+ employees, mature operations	$100K-$2M+, Enterprise Data Partnership listing	Compensation under the agreed data partnership
Mode	Buys a de-identified copy of operating records	20+ U.S. office staff, firms fewer	$100K to $5M by its estimate	Fixed price, no broker fee
Nyne	Licenses workflow data to AI labs	20+ U.S. W-2 staff, 3+ years	Six figures, by its headline	On acceptance, per agreement
Ooak Data	Licenses anonymized operating data	No minimum, closed companies too	$20K to $300K, by its own page	Stated as 2 to 4 weeks
Polyshares	Licenses operational data, you keep it	Any company with real history	$100K to $2M+	Paid per license, no commission
pre.dev	Licenses private code for AI coding tasks	Agencies and companies with private repos	Example: $6,000 per repo	Net 15 after funds land
Protege	Royalty marketplace, mostly health and media	Must hold the data rights	No range published	Revenue share on qualified deals
Scale AI	Turns workflow records into training data	40+ employees preferred	$10K to $1M+, illustrative	Recurring, terms not published
The General Data Company	Licenses industrial company archives	Industrial firms, closed ones too	No range published	One-off price or revenue share
Turing Lazarus	Acquires rights to operating records	Any company with real history	Up to $1M, one payment	Single payment, no recurring
Marketplaces
Brickroad	Packages data feeds and routes licenses	Any size, first-time licensors too	No range published	Split set in each license
Dappier	Paid AI access to publisher data	Publishers and data vendors	No range published	Per-unit pricing, ZeroClick and Stripe
Datarade	Listing marketplace, does not buy	Data providers who apply	No range published	Provider pays a commission
DataVendor	Self-serve marketplace for private code and data	Organizations, self-serve	You set the price	Sale price less 20% on withdrawal
Magma	Licenses AI agent interaction data	Companies building AI agents	No range published	Revenue share as data licenses
Microsoft Publisher Content Marketplace	Pilot marketplace licensing publisher content	Publishers, pilot stage	No public terms	Paid on delivered value
Opendatabay	Dataset marketplace, also brokers deals	Verified registered businesses	You set the price	Sale price less a 5% to 30% fee
Replay	Licenses de-identified operating data	From about 20 employees	$10K to $1M+ by size	Upfront plus revenue share
SuperAnnotate	Prepares and licenses records to AI labs	Any size, any industry	$100K to $1M+, indicative	On delivery and each relicense
TollBit	Charges AI bots for publisher content	Publishers	No range published	Not published
Troveo	Licensing marketplace, business data since 2026	No minimum stated	Six figures and up	Revenue share on each sale
Versos	Prepares video libraries for AI licensing	Video library owners	No terms published	Not published
Shutdown services
SimpleClosure Asset Hub	Licenses or sells startup code and data	Closing or operating companies	$10K to $100K per company	A percentage on a sale or license

AI and data labs

They license or partner for the data themselves.

Appen

An AI training-data company that licenses company workflow records to AI labs and quotes each partnership on its own.

Who they take: No minimum published on the page we read. Companies winding down are asked to flag it on the form for a faster reply.
How they make money: Not published. Appen licenses the data to AI labs and says it covers extraction, anonymization and transfer, with no deductions from what you are paid.
Deal size they state: Quoted per partnership
De-identification: Personal data is detected and removed, with synthetic rewriting where the workflow structure has to survive. No de-identification standard is named.

Full profile Their page

Centific

An AI data company that extracts, anonymizes and licenses company operating records to frontier AI labs, paying on acceptance.

Who they take: No minimum published. Its form's headcount options start at up to 50 employees, and it takes operating, winding-down and closed companies.
How they make money: Not published. Centific licenses the data to AI labs and says it pays for extraction, anonymization and transfer, with no setup fees or deductions.
Deal size they state: Written estimate after first call
De-identification: Personal data is removed, with synthetic rewriting where a record's structure must survive. You review an anonymized sample before use. No de-identification standard is named.

Full profile Their page

Collinear AI

A research lab that licenses code repositories and business-app history to build training and evaluation environments for AI agents.

Who they take: No minimum published. It says the price rises with 15 or more people at peak, several years of history, human-written code and work visible across several tools. It does not want AI-generated code, loose documents or anything you lack the rights to.
How they make money: Not published. It uses the data to build its own agent training and evaluation environments.
Deal size they state: Range quoted after first call
De-identification: Personal data is removed after export: names become consistent pseudonyms and identifiers are stripped. You can hold back anything you prefer not to send. Consistent pseudonyms keep records linkable.

Full profile Their page

Defined.ai

A data lab with a long-running marketplace for licensed datasets. It wants datasets with consent documentation, not whole company archives.

Who they take: No fixed minimum dataset size. Partners need consent and licensing documentation.
How they make money: Licensing fees from marketplace sales. Its cut is not published.
Deal size they state: No range published
De-identification: Requires consent and licensing documentation from the partner and runs an ethics and compliance review before listing.

Full profile Their page

Google (content partnerships)

Google says it is piloting a program to pay a focused set of content providers for non-public content. No public terms or application.

Who they take: By invitation. It engages a focused set of content providers and publishes no criteria.
How they make money: Not applicable. Google uses the content to improve its own AI models and products.
Deal size they state: No public terms
De-identification: Not published.

Their page

Halluminate

A data research lab building AI benchmarks and training environments for finance and consulting work, with an interest form for company data partnerships.

Who they take: Not published.
How they make money: Not published.
Deal size they state: No terms published
De-identification: Not published.

Their page

Handshake AI

A worker-and-employer network that runs an expert fellowship and a separate company-data partnerships program.

Who they take: 20+ full-time employees and 3+ years in operation (referral page).
How they make money: Licenses de-identified company data to AI labs. Its FAQ says payments follow defined delivery milestones, typically 30 to 60 days after it accepts the de-identified dataset. Its margin is not published.
Deal size they state: $100K to $4M+Public figure, see source
De-identification: Its data partnerships page lists de-identification before delivery and never names partners. The same page notes that no de-identification process removes re-identification risk entirely, and that a third party may be able to re-identify personal information by combining it with other data.

Full profile Their page

Labelbox Prism

Labelbox's data-licensing program, which licenses company operating data to frontier AI labs for an upfront fee plus a revenue share.

Who they take: No minimum published. Companies in distress or wound down can license too, with a single payment for the whole operating corpus.
How they make money: Not published. Prism licenses the data to AI labs, and any exclusivity premium is priced to the buyer.
Deal size they state: About $50K to $1M+Public figure, see source
De-identification: Names, emails and internal IDs are tokenized before data leaves, and workflow traces are licensed at team level. No de-identification standard is named.

Full profile Their page

Mercor

An expert-labor platform with a company-data arm that extracts workflow data, says it anonymizes it, and licenses it to AI buyers.

Who they take: None stated on its data page.
How they make money: Licenses datasets it describes as anonymized to AI buyers. Says it covers extraction and anonymization costs and charges the company no fees. Its own margin is not published.
Deal size they state: Up to $300K reportedPublic figure, see source
De-identification: Its data page lists fully automated extraction and anonymization. No de-identification standard is named.

Full profile Their page

micro1

An AI data company that runs company-data partnerships for AI development, alongside an expert-labor marketplace. Its page says companies keep ownership of their underlying data.

Who they take: micro1's current public criteria list 30+ employees, mature operations and primarily English documentation, with US companies prioritized. Acceptance depends on its review and current demand. Confirm exceptions directly with micro1.
How they make money: Runs company-data partnerships for AI development and pays companies under the agreed partnership. Its pages name no AI lab that receives the data.
Deal size they state: $100K-$2M+, Enterprise Data Partnership listingPublic figure, see source
De-identification: Its pages say PII is removed when relevant and data runs through isolated pipelines. In an undated recorded forum talk, micro1's David Revland described transforming data instead of redacting it, with names as consistent stand-ins across tools. micro1 staff said companies cannot pre-scrub. Consistent stand-ins keep the links between records, which is why re-identification is the open question on the risk map. Confirm the current method with micro1.

Full profile Apply to micro1 (paid link)Their own pagePaid link: micro1 may pay us a referral fee if a company we introduce goes ahead with it. You pay nothing extra.

Mode

Buys a de-identified copy of a company's operating records for AI research and training, at a price fixed before export.

Who they take: 20+ full-time U.S. office employees for most businesses and nonprofits, 10+ employees for accounting firms and 6+ for law firms. Several years of owned records and a mainly U.S. team are the strongest fit.
How they make money: Mode buys the agreed copy and resells it for AI research and training. It says it charges you no broker fee.
Deal size they state: $100K to $5M by its estimatePublic figure, see source
De-identification: Sensitive material is removed, and names and contact details are removed or transformed in a secure workspace before onward delivery. No de-identification standard is named.

Full profile Their page

Nyne

Licenses de-identified company workflow data to frontier AI labs for training and evaluation, paying once the dataset is accepted.

Who they take: 20+ full-time U.S. W-2 employees and 3+ years of operational workflows, plus a decision-maker such as the owner or CEO who can approve packages.
How they make money: Not published. Nyne licenses the data to AI labs.
Deal size they state: Six figures, by its headline
De-identification: Nyne strips personal data, passwords and financial records so they cannot be traced to people or accounts. No de-identification standard is named.

Full profile Their page

Ooak Data

A Y Combinator-backed lab that licenses anonymized company operating data as training material for frontier AI models.

Who they take: No minimum published. Its estimator starts at 1 to 10 people and 2+ years of history, and covers operating, winding-down and closed companies.
How they make money: Not published. Ooak licenses the anonymized dataset to AI buyers under a code name.
Deal size they state: $20K to $300K, by its own pagePublic figure, see source
De-identification: One entity dictionary replaces names, companies and addresses the same way across every source, and faces, logos and signatures are blacked out. Consistent stand-ins keep records linkable.

Full profile Their page

Polyshares

A buyer that licenses a company's operational data for a defined scope and term, says it anonymizes it, and supplies AI labs. Its data handling page says buyers see de-identified listings in a buyer portal, and that you keep the business and the underlying data.

Who they take: Any company with real operating history, per its site. It says record length and texture matter more than headcount. No employee minimum stated.
How they make money: Licenses the data, says it anonymizes it, and supplies AI labs. Its site says there is no fee and no commission to the company on a license. It does not say which labs take which data.
Deal size they state: $100K to $2M+Public figure, see source
De-identification: It says an outside partner strips identifying fields, and that roles, timestamps and work content stay. Its data handling page says identifiers are stripped before Polyshares or any lab sees the record, and names two people who can see identity alongside the anonymized data. An article by co-founder Ryan Lock says Polyshares never touches raw files. The same page says listings in its buyer portal carry no company name, website, people or exact locations.

Full profile Their page

pre.dev

A coding-agent company whose labs arm licenses private codebases and turns them into coding tasks for frontier AI labs.

Who they take: No minimum published. Its sign-up offers an agency or company account as well as a personal one. This profile covers the company account.
How they make money: pre.dev Labs builds coding tasks from licensed code and supplies them to AI labs. Labs get tasks, not access to your repository.
Deal size they state: Example: $6,000 per repoPublic figure, see source
De-identification: Not described on the page we read.

Full profile Their page

Protege

A data lab that runs a governed marketplace for licensed datasets, mostly healthcare, video and audio. Its homepage does not mention workplace chat or email.

Who they take: None stated. Companies hold their content or data rights outright or hold broad rights.
How they make money: Revenue share on qualified deals that include your data. The split is not published.
Deal size they state: No range published
De-identification: Its video data-provider page lists automated detection and removal of personal identifiers, sampling and quality assurance, and optional anonymization or voice modification, depending on the dataset and jurisdiction.

Full profile Their page

Scale AI

An AI training-data company that turns de-identified company workflow records into training data for frontier models, with approval on every data package.

Who they take: 40+ employees is its preferred starting point, assessed case by case. It also looks for years of records, real workflows, domain depth and an owner who can approve packages.
How they make money: Not published. Scale says it designs partnerships for recurring revenue.
Deal size they state: $10K to $1M+, illustrativePublic figure, see source
De-identification: Scale anonymizes inside your environment before any data is extracted, stripping personal data so records cannot be traced to people or accounts. No standard is named.

Full profile Their page

The General Data Company

Licenses anonymized operating records of industrial companies to AI labs and to manufacturers training their own models.

Who they take: Industrial companies, including ones no longer trading. It takes only what no longer gives you an edge: current designs and prices stay with you.
How they make money: Not published. It licenses the anonymized results to AI companies.
Deal size they state: No range published
De-identification: Names, addresses and other identifiers are replaced with invented ones, consistently across documents. Its model finds them, a person checks the result, and the key linking new names to real ones is then deleted.

Full profile Their page

Turing Lazarus

Turing's Project Lazarus acquires defined rights to a company's operating record for frontier AI training, for a single payment.

Who they take: Any company with a real operating history and someone authorized to transfer the data. Operating, winding-down and closed companies all qualify.
How they make money: Not published. Turing uses the record for frontier AI training.
Deal size they state: Up to $1M, one paymentPublic figure, see source
De-identification: Personal data and secrets are removed before anything reaches a model. You approve the scope before work begins.

Full profile Their page

Marketplaces

They list or license data to buyers and share the revenue.

Brickroad

A data-feed network that packages a company's data, reviews it and routes licenses to AI labs, funds and corporate buyers. It does not buy the data itself.

Who they take: No minimum. Its FAQ says a small or niche feed is welcome and that it is built for companies licensing data for the first time.
How they make money: Not published. Its FAQ says the split is set per offering, in writing, in the license you approve. Its pricing page lists fees for buyers only.
Deal size they state: No range published
De-identification: Anonymization during enhancement. Connecting requires representations on provenance, rights, material non-public information and personal data. No de-identification standard is named.

Full profile Their page

Dappier

A marketplace of real-time publisher data that AI agents query and pay for, plus AI answer and conversational-ad products for publishers. It serves publisher content, not company archives.

Who they take: Aimed at publishers and data vendors. No eligibility criteria published on the pages we read.
How they make money: Per-unit pricing quoted through its API, paid through ZeroClick and Stripe checkout, and conversational ads. The revenue share to publishers is not published.
Deal size they state: No range published
De-identification: Not stated. It serves publisher content, and company records are not in its catalog.

Full profile Their page

Datarade

A listing marketplace where buyers contact data providers. It does not buy data.

Who they take: None. Providers apply through its provider portal.
How they make money: Commission paid by the provider when a buyer purchases. Its homepage says the service is free for buyers and paid by the provider.
Deal size they state: No range published
De-identification: Not found. It is a marketplace, and the seller handles the data and any anonymization.

Full profile Their page

DataVendor

A self-serve marketplace where organizations list private code, datasets and AI training tasks for invited AI-company buyers.

Who they take: Organizations only. Vendor setup is self-serve with no application review. A higher vendor tier, which can answer buyer requests, needs a signed agreement.
How they make money: A 20% platform fee on withdrawals to a bank. Earnings converted to its own platform credits carry no fee. Buyers pay a separate transaction fee.
Deal size they state: You set the price
De-identification: Not described on the pages we read. Buyers see redacted previews before purchase.

Full profile Their page

Magma

Packages the interaction data that a company's AI agent produces and licenses it to AI labs for a revenue share.

Who they take: Companies building AI agents. No size minimum published.
How they make money: A share of license revenue. The split is not published and sits in the data agreement signed at onboarding.
Deal size they state: No range published
De-identification: You set the scope. Its FAQ says handling and retention of your users' data are covered in the data agreement. No de-identification method is described.

Full profile Their page

Microsoft Publisher Content Marketplace

A pilot marketplace where publishers set terms to license premium content to AI builders, starting with Microsoft Copilot, and are paid on delivered value.

Who they take: Publishers of any size, by its announcement. Participation is voluntary, and the pilot is expanding partner by partner.
How they make money: Not published. AI builders license content through the marketplace.
Deal size they state: No public terms
De-identification: Not published.

Their page

Opendatabay

A UK data marketplace where registered businesses list datasets for AI buyers, with a sliding platform fee on each self-serve sale.

Who they take: Registered businesses, organizations or legal entities only, verified by KYC and KYB checks. You must own the data or hold the right to license it.
How they make money: A platform fee on self-serve sales that slides from 30% on small sales to 5% above $250,000. In brokered deals it takes a license at your price and sets its own price to the buyer.
Deal size they state: You set the price
De-identification: The provider is responsible. Data must be collected lawfully under UK GDPR, GDPR and CCPA. Opendatabay does not describe de-identifying data for you.

Full profile Their page

Replay

Licenses de-identified company operating data to frontier AI and data labs, paying upfront plus a revenue share.

Who they take: The page says any industry, with one requirement: records that document operational processes. Its value bands start at 20 to 50 employees.
How they make money: Not published. Its page says the company gets an upfront payment plus a revenue share of every license.
Deal size they state: $10K to $1M+ by sizePublic figure, see source
De-identification: De-identification before licensing. Its AI buyers page says stable placeholders stay consistent across every system, so the same entity stays linked end to end. Consistent placeholders keep the links between records, which is why re-identification is the question to ask. No outside standard is named.

Full profile Their page

SuperAnnotate

An AI data platform that prepares a company's operational records and licenses them to the AI labs and enterprise AI teams the company approves.

Who they take: Any size, any industry, anywhere, by its FAQ. You need records built up over time and the rights to license them.
How they make money: Not published. It licenses the data to buyers you approve.
Deal size they state: $100K to $1M+, indicativePublic figure, see source
De-identification: Identifying detail is removed to the standard your team sets. Data is kept for an agreed period, then deleted, and deletion is honored on request.

Full profile Their page

TollBit

A platform where publishers meter and charge AI bots and agents for access to their content. Built for publishers' web content, not company archives.

Who they take: Aimed at publishers. The page says the core platform is free for publishers.
How they make money: No commission or fee percentage is published. Its Publisher Platform Agreement says publishers pay for optional extra services at prices set in their account portal.
Deal size they state: No range published
De-identification: Not applicable in the way it is for company records. It handles publisher content access and traffic analytics, and names no method for personal data.

Full profile Their page

Troveo

A licensing marketplace that started with video and audio and added business data in 2026.

Who they take: No minimum stated on the business-data page we read. Its estimator's headcount options start at 20, a form setting. Its guide says a 40-person firm with specialized records can be worth more than a 4,000-person firm with generic data.
How they make money: Revenue share on every sale to its buyers. The split is not published.
Deal size they state: Six figures and up
De-identification: Anonymization by scrubbing, to a standard it calls documented. Its guide says models already trained survive the end of a license, because nothing can be untrained.

Full profile Their page

Versos

A platform that indexes studio and creator video libraries in their own cloud storage and delivers licensed datasets to AI companies.

Who they take: Not published.
How they make money: Not published.
Deal size they state: No terms published
De-identification: Not published.

Their page

Shutdown services

They sell data from companies they are winding down.

SimpleClosure Asset Hub

A company shutdown service whose Asset Hub, launched in Apr 2026, licenses or sells source code and workspace data. It was built for startups winding down and now takes operating companies too.

Who they take: No staff minimum stated. Its AssetHub page says you do not need to be shutting down, and operating companies can explore licensing.
How they make money: Its AssetHub page says there is no upfront cost and SimpleClosure takes a percentage only when assets are sold or licensed. The percentage is not published. Its terms refer to a separate asset purchase agreement.
Deal size they state: $10K to $100K per companyPublic figure, see source
De-identification: Its AssetHub page says every dataset is processed for PII before licensing, and that sensitive employee, compensation and health information is excluded. Its terms say only that its evaluation may include an assessment of embedded personal information. No method or standard is named. Privacy advocates dispute that anonymization of workplace messages works, and Gizmodo reported that a policy group, the Center for AI and Digital Policy, wrote to the US Senate Commerce Committee urging stronger FTC oversight of AI companies.

Full profile Their page

How to compare offers

No company above publishes its full contract, so the first job is to ask for the terms in writing. These twelve questions come from the terms the companies do publish and from the law-firm checklists linked on the risk map. Put the answers side by side before you sign anything, and have counsel read the contract.

Who is the end buyer? Ask whether the company buys the data itself or passes it on, and which buyer category ends up with it. Most companies above name no end buyer.
License or sale? The labels vary. Troveo says you keep ownership and it licenses on your behalf. micro1 says companies keep ownership of their underlying data.
SimpleClosure's terms grant it a perpetual, irrevocable license. Ask counsel how the label interacts with privacy law, because a license for money can still count as a sale.
Scope and term. Which models, which uses (training, evaluation, retrieval), whether successors and affiliates are covered, and for how long.
Exclusivity. Troveo's guide says exclusivity pays a premium and is time-limited. Ask what exclusive means in your contract and what you give up.
Sample approval. Who sees a sample before a deal, and can you refuse a buyer or a use.
De-identification. Who does it, to what written standard, whether an independent party certifies it, and whether the method keeps consistent stand-ins for names. Consistent stand-ins keep records linkable.
Deletion. Several companies say deleting your data does not undo training that already happened. Ask what is deleted, when, and what proof you get.
Fees and revenue share. The percentage, what counts as net sales, whether the company can claw back a payment, and when you are paid. Mercor states 2 to 4 weeks after the signed agreement. Most others state nothing.
Your warranties and indemnity. Ask for the list. SimpleClosure's terms put six representations and a broad indemnity on the company licensing the data while capping its own liability at the greater of six months of amounts paid or $100.
Liability cap and governing law. Where disputes go, and whether data-protection and IP indemnities sit outside any cap.
Who holds your raw data, and how safely. In every model above, some company holds the raw data at some point. Ask for the security report itself, with scope, auditor and audit dates, and not a badge on a web page.
Audit rights. For recurring or revenue-share deals, ask how you check what the buyer does with the data and what the company reports to you.

Report an error or a missing company

Every entry is a reading of the company's own pages. If a company changed its terms, if we misread something, or if a company belongs here and does not appear, tell us. We correct the entry and note the fix.

Send a correction through the contact form

Where to go next

If you want a neutral second pair of eyes before you talk to any of these companies, take the free fit check or run the seven-question rights check first. The rights check shows where your data could be blocked before any buyer sees it.