| Quick facts | |
|---|---|
| Can a private company do this? | Yes — buyers ran published programs for it throughout 2026 |
| What is published on price | $100K–$500K entry band, per one buyer’s own page |
| Who qualifies | Per that buyer: 30+ full-time employees and several years of records |
| What you keep | The originals, the customers and the company — a license is a copy with defined rights |
| The step most sellers skip | Confirming you have the right to license each record category |
| Biggest avoidable mistake | Signing a perpetual, exclusive, whole-corpus license on the buyer’s first draft |
Section 1
The short answer
A private company can now sell — more precisely, license — a copy of its internal operating records to firms that supply AI model developers. The buyers want the material that never reached the public internet: tickets, work orders, estimates, SOPs, CRM histories, internal messages. They use it to build training and evaluation environments in which AI agents practise real work.
The process has seven steps, covered below: inventory what you hold, confirm what you have the right to license, de-identify, get more than one written quote, negotiate scope and terms, deliver against objective acceptance criteria, and keep the file. Most sellers who go direct do steps four through six only, on the buyer’s paper. The steps they skip are the ones that carry the risk.
Section 2
Which companies buy business data for AI
Business Insider describes the buyers as “a new class of fast-moving data brokers” that acquire companies’ internal data and package it for model developers.[3] That is the accurate picture: the party across the table is usually an intermediary, not the AI lab itself.
| Buyer | What is publicly known | Published pricing | Source |
|---|---|---|---|
| micro1 | Company Data Partnerships; a stated $1 billion, twelve-month budget financed by Citi and Hercules Capital | Publishes bands of $100K–$500K, $500K–$3M and $3M+; says it looks for 30+ full-time employees | Company announcement and program page |
| Mercor | Runs an enterprise workflow-data program; reported as a competing bidder for at least one seller | No public price list that we could verify | Business Insider, October 2026 |
| HUMXN Research Labs | Targets skilled trades, farms and manufacturers; announced a plan to spend $50 million | About $20 per hour for narrated wearable-camera footage; the reporting notes it is unclear where the $50 million will come from | Homepros, October 2026 |
| AI model developers directly | Public-company licensors such as Reddit and Shutterstock contract with model developers themselves | Enterprise-scale agreements; not a realistic direct channel for a single private operator | Reddit and Shutterstock 10-Q filings |
micro1 is the best-documented. On 9 October 2026 it announced a commitment to spend $1 billion over twelve months through its Company Data Partnerships program, financed with capital provided by Citi and Hercules Capital.[2] It describes three ways of working with model developers: licensing transformed raw data, placing an in-progress agent inside a company for feedback, and freezing a company’s data at a timestamp to provide a simulated environment.[3]
In home services the outreach is already direct. Homepros reported in October 2026 that data firms were approaching HVAC and plumbing contractors for service records, job-site photos, equipment manuals and training videos.[5] The home services data licensing hub covers that trade by trade.
Section 3
What AI companies pay for company data
There is one published price list. micro1’s program page states three bands:
- $100K–$500K — qualified enterprise data partnerships
- $500K–$3M — large-scale operational datasets or ongoing participation across multiple teams and business functions
- $3M+ — highly unique, proprietary operational data with significant value for frontier AI development
It says compensation depends on “the quality, uniqueness, and value of the data provided” and that it pays paid in full within 30 days of acceptance.[1] Business Insider reported the company’s range as $100,000 to $2 million for a single midsize company’s data.[3]
The one contractor example on the public record is smaller and differently shaped than its headline suggests. The Information reported in July 2026[4] — and Homepros restated in October — that micro1 pays Michigan HVAC company Vitality Mechanical a reported $1,000 per “labeling task” to document how it handles vendor invoices, credit applications, and payroll, with the company expecting to earn over $150,000 in a quarter.[5] That is recurring payment for task work on back-office workflows, not a one-time price for a file of service records.
For the factors that move a quote inside a band, see how much company data is worth.
Section 4
Who qualifies
By micro1’s published criteria: established companies with 30+ full-time employees, mature business operations, and substantial historical operational data, primarily in English, with US companies prioritised. On volume it says there is no fixed minimum; companies with several years of records across multiple business systems are generally a stronger fit.[1]
In practice three things decide whether a file is interesting, and none of them is size in gigabytes:
- Depth. Several years of continuous history in the same systems.
- Linkage. The same piece of work visible across systems — the estimate, the work order, the invoice and the payment. Business Insider quotes the buyer’s revenue chief: “The connection, the diversity, the traceability of all of these data points, they matter a lot.”[3]
- Rights. Whether you can actually license it, which is the next section.
The record categories themselves are listed on what data AI companies want.
Section 5
How to sell company data, in seven steps
- Inventory what you hold. List every system, the years of history in each, and rough record counts. Metadata only; no records leave at this stage.
- Confirm what you may license. For each category, check four things: your customer-facing privacy policy, your customer contracts, your employee agreements and handbook, and the terms of the software the records live in. This is where most of the scope gets cut. If you run a field service platform, start with what its terms permit.
- De-identify, and document how. Decide what is removed, generalised or excluded, and write the methodology down while you do it. A record you cannot reconstruct later is a representation you cannot give later. See de-identifying business data before licensing.
- Get more than one written quote. A band on a website is not a bid. A bid is a written number, on named record categories, from a buyer that has seen the metadata. One seller in Business Insider’s reporting had offers from two buyers.[3] That is the position to be in.
- Negotiate scope and terms, not only price. The table in the next section lists the terms that matter. Your counsel reads the license.
- Deliver against objective acceptance criteria. Payment that depends on the buyer’s satisfaction is payment at the buyer’s option. Agree what “complete” means before export.
- Keep the file. Preserve the terms you relied on, the consent basis, the de-identification methodology, the scope delivered and the dates. Your future acquirer’s counsel will ask for it.
The same sequence, with timing, is laid out visually in the AI data licensing roadmap.
Section 6
The terms that matter more than the price
| Term | What a buyer’s first draft tends to ask for | What a seller should aim for |
|---|---|---|
| Scope | The entire data corpus | Named record categories you have confirmed you can license |
| Exclusivity | Exclusive, for a premium | Non-exclusive unless the premium is large and no sale is planned |
| Term | Perpetual | Fixed term; rights end or are renegotiated |
| Permitted use | Any use, including resale and sublicensing | Defined uses; resale and sublicensing listed and limited |
| De-identification | Performed by the buyer after delivery | Performed and documented before data leaves, or contractually specified with audit rights |
| Indemnity | Seller warrants it holds all rights | Warranty limited to your knowledge and to the categories reviewed |
| Payment | After acceptance, at buyer’s discretion | Objective acceptance criteria and a payment date |
Scope is the first fight. Business Insider reports that micro1 “it usually wants to license a company’s entire data corpus.”[3] A whole-corpus grant sweeps in customer personal information, employee messages and vendor-governed records in a single signature. Buyer terms do offer protections — micro1’s page describes “a commitment not to resell or distribute your data until personal and identifying information has been anonymized” and states that companies keep ownership of the underlying data[1] — but a marketing page is not the contract. What the signed license says is what binds.
Exclusivity is the second. It is offered as a premium and it is the term most likely to cost you in a later sale of the business, for the reasons set out in data licensing vs. selling your business.
Section 7
Is it legal?
Licensing your own operating records is not unlawful in itself. The exposure sits in specific contents of those records, and the governing rules are identifiable:
- Your own promises. FTC staff have warned that quietly changing terms of service or a privacy policy to permit new data uses could be an unfair or deceptive practice.[9] What your policy said when the data was collected matters.
- State privacy law. California defines “sale” of personal information broadly and sets a statutory standard for “deidentified” data.[10]
- Recorded calls. Consent rules differ by state and voiceprints can be biometric data. See call recording data licensing.
- Third-party terms. Records inside a vendor’s platform are subject to that vendor’s terms of use.
Buyers have their own reasons to care. Since 2 August 2025 the EU AI Act has required providers of general-purpose AI models to publish a summary of the content used for training and to maintain a copyright-compliance policy[8], and the U.S. Copyright Office’s 2025 report on generative AI training examines fair use and the developing market for licensed training data.[11] Documented rights are worth more to a buyer than undocumented ones. The full list is on the legal issues page. None of this is legal advice.
Section 8
The case against, in the sources’ own words
A guide that only tells you why to sell is a sales page. The public record contains real reasons for caution.
- The market may be short-lived. A founder on the buying side told Homepros it “will probably be ultra-saturated in like a year and a half, and then once they get enough data, it’ll be worthless.”[5]
- Licensing revenue is concentrated even for the largest licensor. Reddit tells investors that “substantially all of the contract value associated with our licensing revenue is derived from two of our partners. Our agreements with those partners are approaching the end of their respective current terms.” and that licensing content for AI training is “a novel business model without an established track record.”[6]
- Commodity data is already repricing. Shutterstock’s data offering fell 91% year over year in the third quarter of 2026.[7]
- You may be underpaid. Homepros relays an economist’s view that buyers are not paying close to what the data is worth, because it trains systems that persist after the payment.[5]
- Your employees have a stake and no say. Business Insider notes that current and former employees are unlikely to receive compensation or a voice, and quotes a law professor calling such sales an “ethically gray area.”[3]
Frequently asked questions
Can I really sell my company’s data to AI companies?
How much do AI companies pay for company data?
Which companies buy business data for AI training?
What kind of company qualifies?
Is it legal to sell company data to AI companies?
Do I lose ownership of the data?
Will selling data affect a future sale of my business?
How long will this market last?
Do I need an advisor or broker to sell company data?
References
Sourced from primary industry reports, government data, SEC filings, and peer-reviewed valuation research. Retrieved at time of publication.
- 1.micro1 — Monetize your company’s operational data — Company Data Partnerships program page — A buyer’s own published payout bands, eligibility criteria, payment terms and data-handling commitments; marketing claims, not audited outcomes (read November 2026)
- 2.micro1 — micro1 announces commitment to spend $1B on enterprise data acquisition over the next 12 months — Announcement of 9 October 2026, financed with capital from Citi and Hercules Capital; also issued via Business Wire (2026)
- 3.Business Insider — Micro1 is making a $1 billion bet on AI’s latest data craze — Seller interviews, the range micro1 says it pays, and a Gartner estimate of AI data spending, 26 October 2026 (2026)
- 4.The Information — Why Data Labeling Startups Are Paying HVAC Companies $150,000 For Data — Original report of the HVAC example, by Stephanie Palazzolo, 29 July 2026 (subscription required) (2026)
- 5.Homepros — Tech firms’ hunt for data reaches HVAC, plumbing contractors — Trade-press report of 17 October 2026 restating the HVAC example and quoting a buyer-side founder on how long the market may last (2026)
- 6.Reddit, Inc. — Form 10-Q for the quarterly period ended September 30, 2026 — Content licensing revenue, partner concentration, and the scraping risk factor (2026)
- 7.Shutterstock, Inc. — Form 10-Q for the quarterly period ended September 30, 2026 — Data, Distribution and Services segment revenue and the decline in the metadata data offering (2026)
- 8.European Union — Regulation (EU) 2024/1689 (Artificial Intelligence Act), Article 53 — Obligations for providers of general-purpose AI models — Copyright-compliance policy and public training-content summary; in force 2 August 2025 under Article 113(b) (2024)
- 9.U.S. Federal Trade Commission — AI (and other) Companies: Quietly Changing Your Terms of Service Could Be Unfair or Deceptive — FTC staff position that retroactively broadening data-use terms can be an unfair or deceptive practice (2024)
- 10.State of California — California Consumer Privacy Act, Civil Code § 1798.140 — definitions of "deidentified" and "sale" — Statutory standard a de-identified operational record must meet, and the breadth of "sale"
- 11.U.S. Copyright Office — Copyright and Artificial Intelligence, Part 3: Generative AI Training (pre-publication version) — The Office’s analysis of training on protected works, fair use, and voluntary licensing of training data (2025)