How data is priced

Data is usually priced in one of seven ways: a one-off sale, a subscription, a price per use, a revenue share, a price tied to the buyer's results, a price hidden inside another product, or an exchange. The choice of model matters as much as the number, because it decides whether the seller is paid once or keeps earning as the buyer uses the data.

In short. There are three ways to set the level of a price: from what the data cost to produce, from what similar data sells for, or from what it is worth to the buyer. The first two are the easiest to use; the third is the only one that tracks what the data is worth. The model then decides how that price is collected over time. Around both sit the terms that move the price: exclusivity, freshness, coverage, granularity, permitted use, term and the number of users.

How do you set the level of a data price?

Three anchors, in rising order of difficulty.

Cost. What it took to collect, clean and deliver the data, plus a margin. It is easy to compute and often too low, because data costs little to copy once it exists. A price built on cost says nothing about what the data does for the buyer.

The market. What comparable data sells for. Useful where a market exists, such as market data, credit data or audience data, and a weak guide where the data is unique, because then there is nothing to compare it with.

Value to the buyer. What the buyer gains from the decision the data improves: the extra sales, the lower losses, the faster trial. This is the anchor that tracks what the data is worth, and the hardest to use, because the seller has to understand the buyer's decision well enough to estimate the gain.

What pricing models do data sellers use?

1. One-off sale. The buyer pays once for a defined dataset, often with a perpetual licence to use it. Simple to agree and to account for, and it ends the income from that buyer for that use at the moment of sale; the seller can still sell the same data to others unless the contract says otherwise. OneMedNet agreed in 2026 to supply de-identified medical imaging for AI model training in a deal worth more than $11.5 million, under a fee that does not share in what the trained model later earns.

2. Subscription. The buyer pays a recurring fee, usually annual, for access to data that keeps being refreshed. This is the standard model for market data, company data and benchmarks, and the one that values freshness. Subscriptions are usually tiered: by the number of users or seats, by the fields and history included, by how often the data refreshes, by geography. Pyth Network launched Pyth Pro in September 2025 as a subscription service covering cryptocurrencies, equities, fixed income, commodities and foreign exchange. Large data licences are often reported as total contract value: Tempus AI reported about $200 million of data-licensing contract value booked in one quarter of 2026, and named five drugmakers with data licensing agreements worth more than $100 million each.

3. Price per use. The buyer pays for each call, query, record or download. It lowers the barrier to a first purchase and lets revenue grow with use, and it needs metering the seller can trust. Redpine, which gives AI tools access to scientific papers, listed Redpine Science at $1 per 1,000 tokens with no upfront subscription, and each participating publisher sets its own per-transaction price.

4. Revenue share. The seller takes a percentage of what the buyer earns from a product that uses the data. It grows with the buyer's success, and it is only worth what the seller can verify. The News/Media Alliance's March 2026 agreement with Bria pays participating publishers a recurring share of revenue, distributed according to how much each publisher's content is used. Revenue shares also pay distributors: IEX made data vendors eligible for an incentive tied to the new subscriptions they bring to its market data products.

5. Price tied to the buyer's results. The seller is paid in part when the buyer's outcome happens: a milestone reached, a product sold, a sale attributed to an advert. It is the model closest to value, and it needs an outcome both sides agree how to measure. Drug discovery collaborations built on data and models often use it: in AbbVie's 2026 collaboration with Iambic, Iambic receives an upfront payment and is eligible for success-based milestone payments and tiered royalties on net sales of products the collaboration generates.

6. Price hidden inside another product. The data is never sold as data. It is priced as advertising, as software, or as a feature of a service the buyer already pays for. Citi launched its commerce media business in September 2026 on its card transaction data: brands buy placements, not the transaction file. Retail and commerce media networks work this way, as do many software products built on data.

7. Exchange. Data for data, or data for a service: two companies pool records to fill each other's gaps, or a company supplies data to a partner in return for analytics, a discount or distribution. No separate payment for the data appears on an invoice, so the price is easy to forget, and it is still a price.

Many contracts combine models: an upfront fee and a revenue share, a subscription with a usage overage, a minimum guarantee against a share.

What moves a data price up or down?

The same dataset can be worth very different amounts under different terms. The terms that move the price most:

• Exclusivity. Data sold to one buyer only is worth more to that buyer, and closes off other sales within the scope and term of the exclusivity.
• Freshness. How often the data refreshes and how old it is on arrival. For many buyers yesterday's data is worth a fraction of today's.
• Coverage and history. How much of the market, the population or the period the data covers.
• Granularity. Record-level data, aggregates or benchmarks. Finer data is usually worth more and carries more privacy obligations.
• Permitted use. Internal analysis, resale, use inside the buyer's own products, or AI model training. Each wider right is worth more, and some, such as training a model, are hard to take back once used.
• Term. How long the licence runs, and what happens to the data when it ends.
• Users. How many people, systems or affiliates may use it.

How do you choose a pricing model?

• The buyer needs the data once, for a project or to train a model: a one-off sale, or a licence with use limits.
• The buyer relies on the data continuously and values freshness: a subscription, usually tiered.
• The buyer uses the data unevenly, or is new and testing it: a price per use.
• The buyer builds a product on the data that earns revenue the seller can see: a revenue share.
• The buyer makes a measurable decision the data improves: a price tied to the result.
• The buyer wants the effect of the data, not the data: the data inside another product.

Two questions help. Can the seller see how the data is used, and can both sides measure what it changes? Where neither is true, the seller is left with a fixed price. Where both are true, the price can follow the value.

What mistakes do data sellers make on price?

• Pricing on cost. Data costs little to deliver once collected, so a cost price leaves most of the value with the buyer.
• One price for every use. A licence that does not distinguish internal analysis from resale or model training risks pricing the narrowest use while granting the widest.
• A share the seller cannot check. A revenue share without reporting or audit rights is a share of whatever the buyer reports.
• Forgetting that the model outlives the number. A one-off sale, however large, ends the income from that buyer. The model decides whether the seller is still being paid when the data proves its worth.

Which share of the value a seller keeps, and why, is the subject of Who captures the value.

Questions people ask

How much is data worth? As much as the decision it improves is worth to the buyer, less what the buyer would pay to get the same answer another way. Cost and comparable prices set a floor and a reference; value to the buyer sets the ceiling.

What are the main data pricing models? One-off sale, subscription, price per use, revenue share, price tied to the buyer's results, data priced inside another product, and exchange. Many contracts combine two or more.

Should data be priced as a subscription or per use? A subscription suits buyers who rely on the data continuously; price per use suits buyers who use it unevenly or are trying it for the first time.

How is data for AI training priced? Often as one-off licences or multi-year access fees for a defined corpus, sometimes with a recurring share tied to use. The permitted use matters more than for most data, because a model trained on the data keeps what it learned.

What makes a dataset more valuable? Exclusivity, freshness, coverage, granularity and the uses the buyer may make of it, and above all how much the buyer's decision improves with it.

Where can I see how my business stands? The Data Monetisation Readiness Assessment on Monetaise is a structured first look.

---

Every deal cited here is in the register, each with its reading.