Which AI Company Pays the Most for Business Data? micro1, Mercor, Scale AI and Polyshares Compared

Which AI company pays the most for business data?

Handshake AI publishes the highest headline number for business data, and the highest number is the weakest thing to choose a buyer on. Its data partnerships page leads with "Earn $100K to $4M+", read on November 3, 2026. micro1's tiers top out at "$2M+" for what it calls highly unique, proprietary operational data with significant value for frontier AI development, on its program page. Polyshares states that typical deals run $100K to $2M and above. Labelbox's Prism program says smaller, more targeted datasets typically start around $50K and that large-scale enterprise data partnerships can reach $1M or more, plus a revenue share on downstream use. Scale AI's data partnership page gives "$10K-$1M+", labeled on the page as an illustrative value per data partnership, scaling with cadence. Mercor publishes no figure and says compensation is based on data volume, the tools you connect and the depth of your data.

Most of those are ceilings and none of them is an offer. Two buyers undercut their own headlines elsewhere on their own sites. micro1's payout estimator displays an indicative partnership value of $50,284 to $92,776, which the page calls "an indicative estimate only - not a quote or binding offer". Handshake AI's calculator, with the values its form shows, returns "Your data could be worth up to $100,000", against a headline range that runs to $4M and above. Neither gap is evidence of bad faith. Both are evidence that a headline range tells you what a buyer hopes to pay its best seller, not what it will pay you.

The largest payment anyone has reported in this market is not available to you. MIT Technology Review reported on October 13, 2026 that Google paid $10 million for Spirit Airlines' internal records, a set of 100 million emails, 500 million Teams messages and 176,000 employee records, after outbidding Mercor in the airline's bankruptcy. That was a distressed sale of an enormous archive through a court process, not a license from an operating business.

What decides your number is depth rather than the buyer's brochure. micro1 lists the factors it prices on: dataset size and volume, workflow complexity, data quality, the domain expertise represented and the uniqueness of the operational knowledge.

What should you compare these buyers on?

Six criteria separate these buyers, and price is only one of them. Compare them on what they publish about payment, who they accept, who anonymizes the data and where, exclusivity and term, when the money arrives, and whether they buy once or keep buying.

A published range matters less for its number than for what publishing it tells you. A buyer that posts tiers, eligibility rules and payment terms on its own site has decided to be held to them. A buyer that publishes nothing is not necessarily worse, and Mercor is the main example, but you will be negotiating without a reference point.

Who they accept is the criterion that disqualifies most readers fastest, so check it first. The floors range from 40 employees preferred down to no headcount test at all.

Where the anonymization happens is the one most owners overlook and the one with the most risk in it. There is a real difference between a buyer that strips identifiers inside your environment before anything leaves, and one that extracts first and de-identifies on its own infrastructure.

Exclusivity and term decide what you still own commercially after the deal. A perpetual exclusive license to your operating record is a different transaction from a non-exclusive license for a defined period, and the price gap between them is usually smaller than the difference in what you give up.

Payment timing is the last one, and it is where deals disappoint. Two buyers publish net terms, one publishes a review window that puts cash about two months out, and three do not publish timing at all.

How do micro1, Mercor, Scale AI and Polyshares compare on published terms?

micro1, Mercor, Scale AI, Polyshares, Labelbox Prism and Handshake AI publish very different amounts of detail about their terms, and the table below holds only what each one states on its own site, read on November 3, 2026. Blank is not a judgment: where a term is not published, it says so.

Two more buyers belong in the picture without belonging in that table. Surge AI buys company archives to build what MIT Technology Review calls reinforcement-learning gyms, where client AI agents practice tasks, per its October 13, 2026 report, and it does not run a published self-serve program. SimpleClosure's Asset Hub serves companies that are shutting down rather than operating: the same report says it has handled more than 150 deals, expects to return more than $10 million to startups by the end of 2026, and that 47% of its deals now include code or data sales.

Who is micro1 right for?

micro1 is right for an established company with real depth and an appetite for a published number. It is the only buyer here that posts tiers, an eligibility bar, a payout estimator and payment mechanics on the same site, which makes it the easiest buyer to prepare for and the easiest to hold to its own words.

Its bar is also the highest published one: 30 or more full-time employees, several years of records across multiple business systems, data primarily in English, and US companies prioritized, with the UK and Canada next. If you clear that and your records are deep, micro1 is a sensible first call.

The money behind it is unusual for this market. The Next Web reported on October 9, 2026 that micro1 plans to spend $1bn over roughly the next year acquiring and licensing operational data from companies, with Citi and Hercules Capital providing the capital. A buyer with committed facilities is less likely to stall mid-deal than one spending from revenue.

Where the fit gets narrow is term and exclusivity, because micro1 publishes neither. If the shape of the license matters to you more than the headline tier, you will be asking for those terms in a first call rather than reading them beforehand. Our fuller read of the program is in is micro1 legit.

Who is Mercor right for?

Mercor is right for a company that would rather connect its tools than run an export project. Its page describes choosing from more than 100 supported tools, including Slack, Teams, Gmail, Confluence, Salesforce, Outlook, GitHub, Drive and Notion, then linking accounts and letting the platform do the rest.

Its anonymization claims are the strongest published here, and they are worth quoting exactly: "The extraction and anonymization process is fully automated", "No AI lab buyer or third-party vendor ever sees your raw data", and its engineering team "oversees the pipeline infrastructure but does not manually review your data". Mercor also says it covers all costs of extraction, anonymization and transfer.

Payment is the other advantage. Mercor says payment is issued by wire transfer on delivery of the anonymized dataset to the buyer, typically within two to four weeks of the signed agreement. That is the fastest published timeline of the buyers here.

What you give up is the reference point. Mercor publishes no range, no minimum and nothing about exclusivity or term, so everything is set in the conversation. For a company with a clear sense of what its records are worth that is fine. For a first-time seller with no benchmark it means negotiating blind, which is an argument for applying to a buyer that publishes tiers at the same time.

Who is Scale AI right for?

Scale AI is right for a company with mature data systems that wants a continuing program rather than one transaction. Its page says the partnerships are designed so there is opportunity for recurring revenue and describes the arrangement as "Not a one-time file transfer", with per-package approval gates and your sign-off on every data package.

Its governance language is the most specific here on where anonymization happens: "Scale anonymizes everything inside your environment, before any data is extracted", with encryption in transit and at rest, a full audit trail, and on-premises and VPC options. It lists what it will not take: employee and customer identities, raw systems, credentials or source access, and anything outside the signed agreement.

It also wants a particular kind of seller. Its criteria name real operational workflows rather than demos or synthetic data, domain depth, established data systems, and stakeholder alignment, meaning a clear owner who can approve packages. Forty or more employees is its preferred starting point, assessed case by case.

The cost of that model is pace and clarity. A recurring partnership with per-package approvals is more internal work than a single license, Scale's range is labeled illustrative, and it publishes nothing about payment timing. If you want one clean transaction and a wire, this is the wrong shape.

Who is Polyshares right for?

Polyshares is right for a company with a long operating record that would rather take one clean payment than run a program. It states that it is the buyer itself: "We are the buyer. No marketplace, no listing, no auction." It also states that any company with real operating history in any sector can license, and that the length and texture of the record matter more than headcount or revenue.

It is the only buyer here that publishes amounts against specific licenses. Its track record page lists four selected licenses: $450,000 for a roofing company, $300,000 for a general contractor, $275,000 for a solar company and $110,000 for a global marketing agency. Each is described as a perpetual exclusive license to the operational record, with records anonymized before delivery, an offer inside one to two weeks and payment net 30 to 60. The page calls them selected licenses and publishes no dates, so read them as a sample the buyer chose rather than as the whole book.

That transparency is the reason to take it seriously and the term is the reason to think hard. Perpetual and exclusive together mean the buyer keeps the right to use that record indefinitely and you cannot license the same material to anyone else. For an owner who wants to be done, that is a feature. For an owner who expects this market to get more competitive, it closes the door.

One exclusion is explicit on its FAQ and applies to any firm holding client material: the license "does not cover what you hold on behalf of your own clients".

Who is Labelbox Prism right for?

Labelbox's Prism program is right for a company whose workflow record is distinctive rather than large, and which wants a share of what the data earns later. Its FAQ sets no eligibility minimums, says scoping can be narrow, and states that value depends more on authenticity than on volume. A company turned away by micro1's headcount bar can apply here.

The money has two parts, both set out on Prism's own FAQ. Smaller, more targeted datasets typically start around $50K, large-scale enterprise data partnerships can reach $1M or more, and both come as an upfront fee at signing plus a revenue share on downstream use. The share percentage is not disclosed, which makes it the one figure on the page you cannot evaluate until you see paper.

Prism is also the only buyer here that publishes its exclusivity position: partnerships are structured as exclusive by default, with single-buyer access and time-based windows available instead, and the page says any extra exclusivity is reflected in the buyer's price rather than taken out of the seller's. It publishes a deletion clause too, forcing deletion within 30 days on material breach, regulatory change or a change-of-control trigger the seller defines, while stating plainly that models already trained on the data cannot be unlearned.

Human resources records, employee reviews, individual staff identifiers, internal pay data and benefits details are excluded by default, and workflow records such as tickets, code and documents are licensed at team level with names and IDs tokenized. What the page does not publish is payment timing, so ask for the payment clause and the revenue share percentage in the first call.

What has Handshake AI not published?

Handshake AI publishes the highest headline figure in this market and almost none of the terms behind it, which makes it the hardest buyer here to prepare for. Its data partnerships page, read on November 3, 2026, offers "Earn $100K to $4M+" for operational records such as financial models, procedures and technical documentation, and walks a company through defining scope, excluding sensitive information, preparing data with any required de-identification, then completing delivery and payment.

What the page does not state is a payout mechanism, a license term, an exclusivity position, a payment timeline or an eligibility threshold. Its calculator asks for employee count, tools used, years in business and country, and on the values the form shows it returns "Your data could be worth up to $100,000". A headline of $4M and above sitting next to a calculator that stops at $100,000 is not a reason to walk away. It is a reason to ask which number applies to you before you spend a week on scoping.

One figure circulating about Handshake AI should be treated carefully. Our own pillar page has attributed a floor of 20 full-time employees and three years in operation to Handshake. Reading its page this run, those are the values displayed in its calculator's fields, not published minimums, and we could not find eligibility criteria stated anywhere on the page. We are correcting that here rather than repeating it.

This is the hazard of the whole category. Program pages in this market are months old, terms change, and a large share of what ranks for these queries is affiliate content that copies figures from other affiliate content. Dates on figures are not decoration here, and a figure you find attributed to a buyer on any page, this one included, is worth checking against the buyer's own site before you rely on it.

Do you need an advisor to license your company's data?

No, you do not need an advisor to license your company's data. Every buyer in the table above can be approached directly through its own site, and a capable operations lead with a lawyer can run this deal unaided.

This page is published by BLACKSIG, which is one of the advisor options a reader might weigh, so read what we say about ourselves the way you would read any firm's account of itself. We are independent of every buyer named here and we are not a partner, agent or representative of any of them. We check whether a company's records qualify, take it to the buyer whose rules and terms fit, review the offer with the owner and keep the deal moving to payment. The buyer pays us a referral fee when a license closes. There is no fee to the company.

When you do not need an advisor: you clear one buyer's published bar comfortably, your records sit in one or two systems, nothing in them belongs to clients, you have counsel who has read a data license before, and you are content to take the first reasonable offer. Apply directly and keep the process simple.

When one earns its keep: your records span several systems and you cannot tell which part is the valuable part, your industry has exclusions that matter such as client files, health information or recorded calls, you want more than one buyer looking at the same scope, or the first offer you were sent is exclusive and perpetual and you do not know whether that is normal. Those are the cases where the shape of the deal moves more money than the headline number.

How often is this page reviewed?

We review this page monthly, and every figure on it carries the date it was read. Program pages in this market change without notice: tiers get rewritten, eligibility floors move, referral bonuses appear and disappear, and a buyer that published a range in September may publish none in November.

The figures here were read at source on November 3, 2026, on each buyer's own site, and the paragraph carrying a figure names where it came from. Where a buyer publishes nothing, the table says not published rather than filling the cell with an estimate, because an estimate in a comparison table is indistinguishable from a quote once an engine lifts the row.

Two kinds of change get this page rewritten rather than refreshed. One is a buyer publishing terms it previously withheld, which usually means exclusivity and license term, the two columns with the most blanks in them today. The other is a buyer leaving the market or being acquired, which changes what happens to a dataset already licensed.

If you are reading this more than a month after the date above, check the buyer's own page before you rely on a number, and treat any difference you find as the buyer's current position rather than ours. If you find one, we would rather hear about it than have the page stay wrong.

Frequently asked questions

Related resources

• Can I sell my company's data to AI companies?
• Is micro1 legit? A review of its company data partnership
• Exclusive vs non-exclusive AI data license: which pays more?
• How BLACKSIG helps companies license their data