Skip to content

AI data market

Data platform versus data product investment: who answers for the rows?

By SourceX Editorial · Reviewed by Noah Loul ·

10 sections

Short answer

A data platform ends where the seller starts answering for the rows: granting the license itself, standing behind what the records contain or working to one buyer's specification. Tooling platforms bill for usage and leave rights between holders and users; data products license datasets they add value to; services sell hours. The EU Data Governance Act draws a similar line. Classify each revenue line by what the customer keeps and who answers for wrong rows.

License terms

Data platform

This page covers data platform versus data product investment: who answers for the rows.Illustrative

Key takeaways

  • The platform ends at the first of three lines: the seller grants rights to the rows, answers for their content, or works to one buyer's specification.
  • EU law already separates tools, neutral intermediaries and value-adding licensors, and bars intermediaries from using brokered data for anything beyond making it available to users.
  • Snowflake's 10-K for fiscal 2026 ties the substantial majority of its revenue to compute, storage and data transfer consumed, not to the rows customers bring.
  • Labels drift: ZoomInfo's 2020 10-K paired software with database access, while its 2025 10-K calls its subscriptions primarily software applications.
  • Normalize each line by what the customer keeps after the contract ends and by the cost that grows with each sale.

Where does a data platform end and a data product begin?

At the first point where the seller starts answering for the rows. A platform sells access to tooling, storage or a venue and leaves the license to the data between the people who hold it and the people who use it. A data product begins when the seller grants rights to specific records in its own name and stands behind what they contain. A data service begins when the seller's staff do work to one buyer's specification.

The distinction matters for data platform versus data product investment because the three earn money differently. Platform revenue tracks usage and can grow with little labor per customer; product revenue depends on holding rights to a body of records worth buying more than once; service revenue tracks hours. One company can run all three, and the diligence task is to find out which one each revenue line really is, whatever the deck calls it.

Three lines, read in the contract rather than the pitch, do most of the work: the rights line (who signs the license to the rows with the customer), the accountability line (whose warranty responds if a row is wrong or unlicensed) and the labor line (whether delivery needs work shaped to one buyer).

How the EU Data Governance Act sorts tools, intermediaries and data products

EU law already draws a version of the same boundary. The Data Governance Act, Regulation (EU) 2022/868, defines a data intermediation service as one that aims to establish commercial relationships for data sharing between data holders or data subjects and data users. Its recitals say that cloud storage, analytics, data sharing software, browsers and email are not intermediation services when they only supply technical tools and do not aim to set up such relationships, and they name data marketplaces and data pools as services that are.

The definition then leaves out services that obtain data from holders, aggregate, enrich or transform it to add substantial value, and license the resulting data to users without establishing a commercial relationship between holders and users. That is close to a legal description of a data product: the seller takes data in, adds value and licenses the result in its own right.

Neutral intermediaries pay for that status. Article 12 keeps a provider from using the data it brokers for anything except making it available to users, and requires the service to be run by a separate legal person. Inside the EU, a company that wants to be both a neutral marketplace and a builder of its own datasets from the same flows faces a structural split, not just a positioning choice. A November 2025 Commission staff working document records complaints that the Article 12 conditions restrain innovation, so the rule is under discussion.

What three companies' own documents show about the label

Public filings and documentation show the three engines, and how labels drift between them. Snowflake's 10-K for fiscal 2026 says the substantial majority of its revenue comes from fees for the compute, storage and data transfer resources customers consume on its platform: whatever rows pass through, it is paid for usage. Its 10-K for fiscal 2023 describes professional services, mainly consulting, on-site technical solution services and training, as a separate line.

AWS Data Exchange sits between platform and product. A subscriber accepts the provider's offer, including its price, duration, data subscription agreement and refund policy, while AWS invoices subscribers and collects payment on the provider's behalf. The license runs from the provider; the platform runs the venue and the billing.

ZoomInfo shows a label moving. Its 10-K for 2020 described subscriptions as software applications together with related access to its databases; its 10-K for 2025 describes subscription services as primarily software applications, priced by functionality, by the number of users and by the amount of data a customer integrates into its systems. Pricing that rises with the data integrated suggests the records still carry much of the value, which is why a label deserves a test rather than acceptance.

Each document read against the three lines:

What three companies' own documents show about the label
DocumentLabel it usesRevenue engine describedLine to test first
Snowflake 10-K, fiscal 2026PlatformFees for compute, storage and data transfer consumedLabor: the size of the separate professional services line
AWS Data Exchange user guideData exchangeProviders' offers, invoiced and collected by AWS on their behalfRights: each provider's agreement governs the data
ZoomInfo 10-Ks, 2020 and 2025Software applications, earlier with database accessSubscriptions priced partly on the data a customer integratesAccountability: how much of the value is the compiled records
EU Data Governance ActData intermediation serviceMatching data holders with data usersRights, in law: value-adding licensors fall outside

A five-question boundary test for any contract

Run these against the customer contract and the supplier contract, not the website:

  • Who signs the license to the rows with the customer: the company itself, or a third-party provider whose terms the company only attaches?
  • If a row is wrong, unlicensed or carries personal data it should not, whose warranty or indemnity responds?
  • What can the customer still use after the contract ends: nothing, a copy under a continuing license, or only the models and results built during the term?
  • Which cost rises with each new sale: compute, a new dataset to source and prepare, or hours of staff time?
  • Does the company use data passing through its systems for anything beyond delivering it, such as building datasets of its own?

Usage, listings, licenses, hours: four engines on one yardstick

Normalizing by scope means comparing each line on the unit its customer actually buys. Platform revenue per unit of usage, marketplace revenue per listing that sells, product revenue per licensed dataset version and service revenue per hour are not interchangeable, and adding them into one growth rate hides the mix an investor is paying for.

The comparison, with what each line leaves the customer and the unit that makes it comparable:

Usage, listings, licenses, hours: four engines on one yardstick
Line of businessWhat the customer keeps after the contractWho answers for the rowsCost that grows with each saleNormalize revenue by
Tooling platformIts own data; nothing of the seller'sThe customerCompute and supportUsage units consumed
Marketplace or intermediaryWhatever each provider's license grantsEach providerListing review and billingTransaction value kept per listing that sells
Data product, one-time licenseA copy under a continuing license, where grantedThe licensorSourcing and preparing the next datasetRevenue per licensed dataset version
Data product, refresh subscriptionPast deliveries, if the license survivesThe licensor, for each deliveryEach refresh deliveredRevenue per accepted refresh
Data serviceWork product, often assigned to the buyerThe buyer, under its own specificationStaff hoursRevenue per labor hour
A horizontal spectrum from tooling platform through marketplace and data products to data service, with a cut line where the seller grants rights to the rows and a second cut line where staff work to one buyer's specification.
Editorial framework. Where each line of business sits, and the two cut lines that end the platform zone and begin the service zone.

When a data product is really a build: the services objection

The strongest objection to taking any label at face value is that platform or marketplace language can cover economics that are mostly people: implementation, custom data builds, labeling and support sold alongside the software. The test lives in the cost lines.

Filings make the first cut easy when they report the lines apart. Snowflake reports professional services separately from usage revenue, and ZoomInfo recognizes implementation and professional services as they are delivered, apart from subscriptions recognized over the contract term. Where a private company does not separate them, ask for revenue split by usage, license, subscription and hours, with gross margin on each.

The harder case is a data product that has quietly become a service: each dataset is built to one buyer's specification, the buyer owns the result and nothing can be licensed again. The rights line says product, the labor line says service, and the economics follow the labor line. Count how many datasets have been licensed more than once; a company that cannot answer is probably selling builds.

What could move the boundary and break the thesis

Each of these would change which side of the boundary earns the margin:

  • EU rules could loosen. If the Article 12 conditions criticized in the 2025 staff working document are relaxed, marketplaces could build their own data products from the flows they broker and compete with pure licensors.
  • Platforms could integrate upward. A platform that owns the customer relationship and the compute can add first-party datasets cheaply and squeeze independent products sold through it.
  • Products could drift into services. Buyers asking for custom scope, labeling and acceptance support pull every product line toward hours.
  • Labels could keep moving. ZoomInfo's own description changed between its 2020 and 2025 filings, so a trend built on labels rather than revenue lines can mislead.
  • Generated data could lower the value of rows. If synthetic data substitutes for licensed records in a buyer's main use, the product side loses pricing power first, while tooling platforms keep billing for compute.

Which side of the line SourceX sits on, by its own description

By its published pages, SourceX sits on the product side of the rights line, and its preparation work takes it partway across the labor line. It licenses business data directly from companies and licenses it onward to AI developers in its own name, which makes it the licensor of record, structurally different from a venue that matches holders with users; how any EU rule applies to it is a question for counsel, not for this page. Its deal terms make it the owner of the de-identified dataset it licenses, while the supplying company keeps ownership of its original records.

The labor line runs through it as well. Preparation and de-identification are handled by SourceX or its processing partner; buyer datasets start from a request and are sourced from U.S. businesses rather than drawn from stock; and very large deliveries stay in the supplier's storage or travel on encrypted drives rather than being hosted, so it is not a hosting platform either. Its investors page describes one standardized licensing contract reused across deals, the feature that would decide whether work per dataset falls over time.

SourceX has not published revenue by line, labor per dataset or how often a dataset is licensed more than once; company and commercial metrics go to investors directly. Its raise under Rule 506(c) is limited to accredited investors, who request materials through the investors page. A useful diligence question is how its revenue would split across the lines in the comparison above.

Questions for any company that calls itself a data platform

Ask every company in a comparison the same questions, with denominators named, so the labels drop out of the analysis:

  • What share of revenue is usage, license, subscription and hours, and what is gross margin on each?
  • For each dataset you sell, who is the licensor of record, and whose warranty responds to a rights claim?
  • How many datasets have been licensed to more than one customer, and on what terms?
  • What can a customer still use the day after its contract ends?
  • Do you build products of your own from data that passes through your systems, and which contract term allows it?
  • If you operate in the EU as a data intermediary, which legal entity provides the service and what separates it from your other businesses?

How this comparison was built, and its limits

The three lines, the five-question test and the comparison table are an editorial framework built on 27 October 2026 from the Data Governance Act text and a 2025 Commission staff working document on EUR-Lex, Snowflake's 10-Ks for fiscal 2023 and 2026, ZoomInfo's 10-Ks for 2020 and 2025, AWS Data Exchange documentation and SourceX's published pages. The EUR-Lex, SEC and AWS sources were verified through search excerpts from their own domains and should be opened by a reviewer before release.

Nothing was measured. The tables describe structures, not performance; they do not rate Snowflake, AWS or ZoomInfo, and the Data Governance Act applies only to services within its EU scope. No SourceX revenue, margin, customer or growth figures were used.

Frequently asked questions

What is the difference between a data platform and a data product?

A data platform sells access to tooling, storage or a venue, and the license to the data runs between holders and users. A data product is a dataset the seller licenses in its own right and stands behind. The difference shows in who signs the license and whose warranty responds.

Is a data marketplace a platform or a product business?

Mostly a platform for the venue and the billing, with the product sitting with each provider. On AWS Data Exchange, for example, subscribers accept each provider's offer and agreement while AWS bills on the provider's behalf. A marketplace that also licenses its own datasets runs two businesses.

Why does the EU Data Governance Act matter to investors?

Within its EU scope it treats neutral matching of data holders and users as a regulated intermediation service, run by a separate legal person that may not use the data for other purposes. Services that add substantial value and license the result fall outside that definition, so the label a company picks has legal consequences.

How should revenue be compared across platform and product businesses?

On the unit each customer buys: usage for platforms, transaction value kept per listing for marketplaces, revenue per licensed dataset version for products and revenue per hour for services. One blended growth rate hides the mix.

Can a data product business turn into a services business?

Yes, when each dataset is built to one buyer's specification, the buyer owns the result and nothing can be licensed again. The rights may read like a product, but the economics follow the labor.

Sources

  • Regulation (EU) 2022/868 defines data intermediation services, excludes tool-only services and value-adding licensing of resulting data, and under Article 12(a) bars other uses of intermediated data and requires a separate legal person. Source
  • A November 2025 Commission staff working document reports that the Data Governance Act's Article 12 conditions on intermediaries are said to constrain innovation. Source
  • Snowflake's 10-K for fiscal 2026 says the substantial majority of its revenue comes from fees based on the compute, storage and data transfer resources consumed on its platform. Source
  • Snowflake's 10-K for fiscal 2023 says professional services consist primarily of consulting, on-site technical solution services and training related to the platform. Source
  • AWS Data Exchange subscribers accept the provider's offer, including price, duration, data subscription agreement and refund policy. Source
  • AWS invoices AWS Data Exchange subscribers and collects payment on the provider's behalf. Source
  • ZoomInfo's 10-K for 2020 described subscriptions as SaaS applications and related access to its databases. Source
  • ZoomInfo's 10-K for 2025 says subscription services primarily consist of SaaS applications priced on functionality, users and the amount of data integrated, with implementation and professional services recognized as delivered. Source
  • SourceX's investors page says it licenses business data directly from companies and onward to AI developers in its own name, uses one standardized licensing contract, and shares company and commercial metrics with investors directly. Source
  • SourceX's How it works page says the supplier keeps ownership of its original data and SourceX owns the de-identified dataset, subject to the supplier's underlying intellectual property rights. Source

Request SourceX investor materials

SourceX shares company and commercial metrics with investors directly. Any offering is made only to verified accredited investors.

See if your company qualifies