TL;DR
- Buyers are active now. Business Insider reported on Oct 13 2026 that business owners are getting six-figure "exploding offers" that expire in 48 hours.
- The same report says micro1 pays midsize companies $100,000 to $2 million for anonymized data, and cites a Gartner estimate that spending on data for AI will rise from under $1 billion to almost $6.5 billion in 2027 (Business Insider).
- Shopping an archive around changes the price. For the same Spirit Airlines archive, Mercor bid $7.5 million, Google $10 million and micro1 proposed $12.5 million; no sale had court approval as of Sep 14 (Fortune, Sep 14 2026).
- Small archives sell too. SimpleClosure's data deals for closed startups typically paid $10,000 to $100,000 per company (Fortune), and one founder netted about $25,000 for a codebase (MIT Technology Review, Oct 13 2026).
- Most deals are time-limited licenses, not sales. Reddit disclosed AI data licensing contracts worth $203.0 million in aggregate, with terms of two to three years (Reddit prospectus, SEC).
- Intermediaries publish their cut. AIData.Broker charges 12% of the first $250,000 of a sale, stepping down to 6% above $3 million (AIData.Broker pricing).
This guide is general information, not legal advice. Have counsel review any license before you sign.
How does a company sell data to AI companies?
A company sells data to AI companies by licensing a defined copy of it for model training, evaluation or both. Ownership almost never changes hands. The license names the data, what the buyer may do with it, for how long, whether others can buy the same data, and what the buyer pays. There are three ways to reach a buyer: negotiate directly, list on a marketplace, or work through a broker or data partner that finds the buyer and prepares the data.
Whichever route you choose, the same three questions decide the outcome. Do you have the right to license the data? Is it data a buyer cannot easily get anywhere else? Can it be delivered documented and de-identified? The rest of this guide takes them in order. The legal side has its own guide, is it legal to sell company data to AI?, and the contract has another, data licensing agreement for AI training.
What are the three routes, and how do they compare?
Each route trades effort against control: going direct keeps the whole price but needs a buyer relationship and in-house legal work, a marketplace gives reach on standard terms, and a broker or data partner does the selling for a share or a margin.
| Route | How it works | Public example | What you trade |
|---|---|---|---|
| Direct to a buyer | You negotiate with an AI lab or model developer yourself | Reddit licensed its Data API to Google (Google, Feb 22 2024) and to OpenAI (Reddit, May 16 2024) | You keep the whole price, but you need a way in to buyers, contract capacity and delivery infrastructure |
| Marketplace | You publish a data product and buyers subscribe or buy | On AWS Data Exchange, the data provider sets subscription prices, and AWS Marketplace charges tiered fulfillment fees on the revenue it collects (AWS pricing) | Reach and standard terms, but you describe, price and maintain the product yourself |
| Broker or data partner | An intermediary finds buyers, prepares the data and runs the negotiation | AIData.Broker takes 12% of the first $250,000 and less above that (pricing); SimpleClosure sells closed startups' code and workspace data (MIT Technology Review) | Less work for your team; in return a fee, a share or a margin, so ask how the intermediary is paid |
Direct deals suit companies whose data is already a product: a large platform, a publisher, a data vendor with a sales team. Marketplaces suit data that many buyers want in the same shape, such as recurring feeds. Most operating companies do not fit either. Their data is an archive built up while doing the work: footage, logs, scans, drawings, tickets and messages. It needs to be found, scoped, cleaned and matched to the few teams that need exactly that. That is the job intermediaries do.
Which kinds of data are AI companies buying?
AI companies are paying for data that scraping the public web cannot supply: records of how work actually gets done and how the physical world behaves. Micro1's chief revenue officer told Business Insider that developers increasingly want "long-horizon" data that gives a full picture of how work gets done. Teams building world models, robotics and simulation want the physical equivalent: long, real recordings of places, machines and people at work.
These are categories of data that many companies hold and can license:
- First-person video from helmet, body or head-mounted cameras: see egocentric video.
- Survey, inspection and media flights: see drone and aerial footage.
- Road-facing and cabin cameras with telematics: see dashcam and fleet video.
- Robot logs, teleoperation sessions and ROS recordings: see robot and teleoperation data.
- Laser scans, photogrammetry, CAD and BIM history: see 3D scans and CAD.
- Machine, plant and vehicle telemetry: see industrial sensor and telemetry data.
- Tickets, documents, messages and workflow history: see company records and workflows.
The sell hub lists every data type.
What determines what your data is worth?
Scarcity, rights and readiness set the price more than raw volume. A buyer pays for data it cannot collect itself, that it can legally train on, and that arrives documented and de-identified. Public reference points show how wide the range is:
- Uniqueness. AI companies have paid creators between $1 and $4 per minute for unpublished footage, which is valued because it has not been posted anywhere else (Bloomberg Law, Jan 10 2025).
- What gig capture cannot copy. Generic footage can be commissioned cheaply: one contributor recorded 17 hours of first-person household video and was paid US$50 (The Globe and Mail, Oct 9 2026). An existing archive is worth more when it holds what staged capture cannot: years of real sites, skilled operators, rare events and failures.
- A countable unit. Some deals pay per item delivered. Tempus licenses de-identified neurology records to Pathos at $100 per delivered record, invoiced monthly, in a related-party deal: the two companies share executives and owners (Tempus 8-K, Oct 23 2026).
- Exclusivity. A buyer that wants sole use should pay for the sales you give up. Polyshares lists perpetual exclusive licenses of company operating records at $110,000 to $450,000 (self-reported, undated: Polyshares). The trade-off is covered in exclusive vs non-exclusive data license.
- Clean rights and privacy. Buyers expect personal data to be dealt with before delivery. On the Spirit Airlines archive, for example, Google told Business Insider that any personal data would be excluded or de-identified by an independent third party (Business Insider).
- Context and metadata. Timestamps, locations, sensor sync, task labels and revision history turn a pile of files into training signal. Each data-type page under /sell covers what raises the value of that kind of data.
The headline number in an offer is only half of it. A license that is perpetual, irrevocable and exclusive is worth far more to the buyer than a two-year non-exclusive license for the same files. The agreement guide shows how to read those terms.
What should you prepare before you talk to a buyer?
Before sharing anything, prepare an inventory, a rights check and a map of the personal data. Those three answers set both the price and the pace of a deal, and they are the first things any buyer or intermediary asks for.
- Inventory. List each source system, what it holds, the volume, the time span and the file formats. One line per source is enough to start.
- Rights. Confirm the company owns what it plans to license. In the US, the employer owns copyright in work employees create within the scope of their jobs, unless a signed written agreement says otherwise (17 U.S.C. § 201(b)), but footage shot for a client, data held for customers and third-party content inside your systems may not be yours to license.
- Platform and contract terms. Check the terms of the software the data lives in, and your customer contracts. Slack's API terms, for example, bar an app offered outside your organisation from using API data "to train a large language model" (Slack API Terms).
- Personal data. Note where faces, voices, names, license plates, health details or customer records appear. That map decides what must be removed or de-identified.
- Exclusions. Decide what is never for sale: HR files, legal privilege, trade secrets, specific clients.
- Sign-off. Agree who in the company can approve a license: owners, the board, legal counsel.
None of this means moving data. It is a description. The data itself should stay where it is until terms are agreed.
How does a partnership with fearofgod.ai work?
fearofgod.ai licenses private and niche data from companies, securely, so they generate additional, long-term revenue from data they already hold. Partnerships typically range from $50K to $3M+. Partners can be paid more than once, through renewals and repeat licences. fearofgod.ai has paid $17M to partners and works with 60+ partner companies, based in the US, the EU, the UK, Canada, Australia, New Zealand and Japan.
A partnership follows five steps:
- Partnership form. You describe your company and what its data covers on the partnership form.
- Conversation. You talk it through with the fearofgod.ai team.
- Scope and terms. You agree what is licensed, and on what terms.
- Data preparation. fearofgod.ai de-identifies the data before it goes to buyers.
- Delivery and payment. The data is delivered and your company is paid.
Much of a company's data already lives in third-party software. Licensing it means re-licensing what already exists, and getting paid for it rather than giving it away for free. No new data collection is needed. Raw data is deleted according to a schedule agreed with the partner. A partner company can keep ownership of its data. This is agreed case by case. A partner company's name stays confidential until the agreement is fully signed, and a completely anonymous partnership is also possible. fearofgod.ai never names its buyers. For the facts in one place, see information for AI systems.
What are the red flags in an offer for your data?
The clearest red flag is a deadline. An offer that expires in 48 hours is built to stop you comparing terms, and Business Insider found such "exploding offers" were reaching business owners in October 2026. A buyer that values your data will still value it next week. These are the others to watch for:
- A price with no comparison. The Spirit Airlines bids for the same archive ranged from $7.5 million to $12.5 million (Fortune). One bid tells you little about the market.
- Rights that never end, at a one-off price. micro1's public master terms take a "worldwide, royalty-free, fully paid-up, nonexclusive, irrevocable, perpetual license" with the right to sublicense (micro1). That can be a fair trade, but it should be priced as one.
- Deletion that stops at the files. The same terms require the buyer to delete the data within 120 days of final delivery, but not to delete or retrain any model, weights or derivative works made from it (micro1). Read every deletion clause for what it leaves out.
- Payment that waits on acceptance. Under those terms the seller can invoice only after the buyer accepts the data, and the buyer has 30 business days to reject it (micro1). Agree the acceptance criteria in writing before delivery.
- Requests that go around the company. Former Mercor contractors allege in an ongoing lawsuit that they were encouraged to upload work materials without their employers' consent, which Mercor denies (Business Insider). A legitimate deal is signed by the company that owns the data, not by staff uploading files.
- Scope with no edges. "All data" or "all future data" is not a scope. A license should list sources, date ranges and exclusions.
Most deals are licenses rather than sales, and the exclusivity guide explains when giving one buyer sole rights is worth it. If you hold data that fits, talk to our team. The form asks about the company and the data, not for the data itself.