Skip to content

AI data market

Asset or channel? Testing an AI data distribution business strategy

By SourceX Editorial · Reviewed by Noah Loul ·

10 sections

Short answer

In an AI data distribution business strategy, the channel beats the asset when buyers can substitute one dataset for another and opening each new supplier costs more than the asset's scarcity premium. Ownership wins when records are close to unique. Public documents show both forces: LiveRamp reports Data Marketplace revenue net because it facilitates trades, while FactSet's content suppliers can end agreements on notice.

Value drivers

AI data distribution

This page covers asset or channel? Testing an AI data distribution business strategy.Illustrative

Key takeaways

  • Distribution out-earns ownership when the cost of a new supplier relationship exceeds the owner's scarcity rent, the asset's value minus the best substitute's.
  • Buyer onboarding is a per-relationship toll; Microsoft's supplier program requires annual attestation and, for higher-risk processing, independent verification.
  • Compare models per buyer relationship, per dataset and per dollar billed, because net and gross reporters cannot be compared on headline revenue.
  • Owners keep leverage even inside channels: opt-in collective rights, publisher-set terms and notice-period exits all appear in public documents.
  • The first number to request from any AI data distributor is datasets per approved buyer relationship per year.

When does distribution matter more than owning the dataset?

Distribution matters more when three conditions hold together: the buyer can substitute comparable records from several owners, every new supplier relationship carries a fixed cost of onboarding, legal review and contracting, and the same buyer comes back for more datasets. Ownership matters more when the records are close to unique for the buyer's task, because then the buyer has to reach the owner whatever channel sits in between.

For an investor this is the question beneath most AI data pitches. One company says it owns scarce records; another says it is the platform buyers already use. Neither label shows where recurring value lands: with the holder of the rights, or with whoever controls buyer discovery and the procurement path a purchase has to travel.

The sections below give a test an investor can run with two estimates, the buyer-side evidence behind it, five distribution models compared on scope-normalized metrics from their own public documents, the strongest counterargument, the risks, and the questions to ask.

The substitution test: scarcity rent against the cost of a new supplier

Compare two shares of the asset's value to the buyer. The owner can charge at most the scarcity rent, the value of its records minus the value of the buyer's next-best substitute. A channel the buyer has already approved removes the one-time cost of opening a new supplier relationship. When that cost exceeds the rent, distribution out-earns ownership.

Write s for the share of the asset's task value that the best substitute delivers, and t for the cost of a new supplier relationship as a share of the same value, counting the buyer's onboarding and the owner's selling and contracting. Ownership earns more when t is below 1 minus s, distribution earns more when t is above it, and the two tie on the line where t equals 1 minus s.

Illustrative: a unique archive whose best substitute delivers 30 percent of its value, sold at a new-supplier cost of 20 percent, sits on the ownership side, because its scarcity rent is 70 percent. Comparable ticket histories whose substitutes deliver 80 percent, at a 35 percent new-supplier cost, sit on the distribution side: the rent is 20 percent and the channel saves 35. A case at 60 and 40 lies on the line.

Three forces move a case. Repeat buying makes every new direct owner relatively dearer, because another dataset through a relationship the buyer has already approved adds no new onboarding. Rights complexity raises t for everyone, and a channel that standardizes it captures the difference. Better public or synthetic substitutes raise s, which shrinks the owner's rent first. None of this needs revenue figures; it needs two estimates an investor can request and challenge.

A chart with new-supplier cost on the horizontal axis and substitute value on the vertical axis, both as percent of the asset's value, split by the line where cost equals scarcity rent into an ownership side and a distribution side, with three illustrative cases: a unique archive on the ownership side, comparable histories on the distribution side and one case on the line.
The substitution test. Illustrative: case positions are example inputs, not company data.

What a new supplier costs the buyer before any data moves

Every supplier a buyer adds costs a fixed review, and large buyers write that cost into their procurement terms. Microsoft's Supplier Security and Privacy Assurance program requires every supplier that processes personal or Microsoft confidential data under its contract to comply with Microsoft's Supplier Data Protection Requirements, complete an annual self-attestation and, where the processing is higher risk, provide independent verification of compliance.

The cost attaches to the relationship, not to the dataset. A buyer that has onboarded a channel can take its fifth dataset through it without repeating the review, while each new direct owner restarts it. That is the mechanism behind t in the substitution test, and it is why the first diligence number to request is how many datasets a typical buyer takes through one approved relationship in a year.

Rights evidence adds to the same toll. A buyer licensing for AI use needs evidence that the seller can grant what it sells. Through a channel, that evidence has to cover every upstream owner, so a distributor's real asset is a repeatable way of producing it, not a list of catalog entries.

Five ways to control the buyer path, as public documents describe them

Each model controls a different stretch of the buyer path: one license in place of many, a marketplace that settles trades between others, an aggregator that resells content it licenses in, a buyer's own storefront, or a gate at the network edge. In none of them does the distributor need to own the records it moves, and each document shows where the owners keep leverage.

The examples were chosen because their public documents state the mechanism plainly, not because they are the largest or best in any category. The last column names the normalized metric that would show whether distribution is earning the margin:

Five ways to control the buyer path, as public documents describe them
Model and public exampleWho holds the rightsWho does the workWhat the public document showsMetric to request
Collective license: Copyright Clearance CenterParticipating publishers, who opt inCCC licenses and administers on their behalfIn 2024 CCC added AI re-use rights for internal use to its enterprise-wide Annual Copyright License, non-exclusive and annually renewableRights holders covered per buyer license; buyer renewal rate
Net-reported marketplace: LiveRamp Data MarketplaceData providersLiveRamp runs the platform and billingRevenue is reported net because its obligation is to facilitate a transaction between data providers and data buyers; amounts remitted to providers are not revenueGross billings and the retained share; buyers active with more than one provider
Licensing aggregator: FactSetThird-party content suppliers for licensed contentFactSet integrates and delivers to clientsSome content agreements can be ended on one year's notice, and two suppliers each exceeded 10 percent of total data costs in fiscal 2024Top-supplier share of data costs; notice periods; client retention
Buyer-run marketplace: Microsoft Publisher Content MarketplacePublishers, who define licensing and usage termsMicrosoft runs the marketplace; AI builders license content for groundingAnnounced February 2026: publishers paid on delivered value, with usage-based reportingShare of a publisher's licensing revenue that runs through one buyer
Network gate: Cloudflare pay per crawlSite owners, who choose to allow, charge or blockCloudflare records billing events, charges the crawler and pays out to the publisherCrawlers either present payment intent or receive an HTTP 402 Payment Required response with pricingPaying share of crawler requests; owners choosing charge over block

Putting gross and net reporters on one yardstick

Normalize every model to the same three units before comparing them: per buyer relationship, per licensed dataset and per dollar the buyer pays. Headline revenue cannot do it, because LiveRamp's marketplace reports only the portion it keeps, while an aggregator that licenses content in carries what it pays suppliers as a cost, as FactSet's references to its data costs show.

These six definitions travel across all five models. Values for private companies, SourceX among them, are not public and have to be requested:

  • Retained share: revenue kept per dollar billed to buyers. Net reporters need to disclose gross billings; gross reporters need to disclose what passes to rights holders.
  • Datasets per buyer relationship: licensed datasets an onboarded buyer takes through one relationship in a year, reported as a distribution, not an average. This is the most direct measure of distribution value.
  • Rights holders per buyer contract: how many owners one procurement unlocks, the breadth a collective license sells.
  • Top-supplier share of input cost: the share of data or royalty cost owed to the largest rights holder, a direct read of owner leverage.
  • Repeat-buyer share: the share of licensed datasets that go to buyers already onboarded, which shows whether the channel compounds.
  • Fulfillment hours per dataset, split by buyer: hours that recur with every buyer behave like services, whatever the revenue line is called.

When ownership strikes back: what suppliers can do to a distributor

Owners regain the upper hand when their records become the only acceptable source, and a distributor's own filings show how exposed it is. FactSet's annual report for fiscal 2024 says some content agreements can be ended on one year's notice, that some suppliers are also competitors, and that suppliers may restrict use of their content or offer it only at prohibitive cost; two suppliers each accounted for more than 10 percent of its data costs that year.

The newer AI channels are built on owner consent from the start. CCC's AI re-use rights reach only works whose publishers opt in, and in Microsoft's marketplace publishers define the licensing and usage terms. A distributor's moat in these designs is the procurement path and the reporting, not control of the content, and it lasts only while owners find the path cheaper than selling direct.

In the substitution test, this is the case where s falls. When substitutes disappear because a category's best records sit with a few owners, the scarcity rent grows past any channel's saving, and value moves back to the rights holder.

Is a marketplace label hiding a services business?

Sometimes, and the revenue categories are where it shows. LiveRamp's filings describe Marketplace and Other revenue as consisting mainly of revenue sharing from access to data through the Data Marketplace, and from fiscal 2024 the same category also includes professional services and certain usage-based items, so one line mixes settlement income with work done for clients.

AI data distribution invites the same blend. Collecting, cleaning, de-identifying and reformatting records for one buyer is labor that resets with the next buyer, even when the revenue is booked as marketplace or platform income. The test is the last metric above: if fulfillment hours grow with each buyer rather than with each dataset version, the channel is partly a services business, and it should be valued with services margins and services scaling in mind.

What could make the distribution thesis wrong

The thesis fails if buyers or owners can make the new-supplier cost small, or if substitutes stop existing. Each of these risks has a visible signal an investor can track:

  • Buyers build their own channels. A large buyer that runs its own marketplace, as Microsoft has begun to for grounding content, captures the procurement saving itself and leaves intermediaries the thinner work.
  • Owners bypass the channel. Large rights holders with legal and sales staff can afford t on their own, so a distributor's base tilts toward smaller owners with smaller assets.
  • Supplier exit and concentration. Notice-period terms like FactSet's mean a channel's catalog can shrink faster than its buyer list.
  • Standardized procurement lowers t for everyone. Shared contract templates, collective licenses or common security attestations cut the toll a channel saves.
  • Substitutes cut both ways. Better public and synthetic data raise s, which hurts owners first but also lowers the total value that any channel can share in.
  • The estimates are soft. s and t are judgments; a company that supplies its own estimates without a method should get no credit for them.

Questions to put to any AI data distributor, SourceX included

By its published description, SourceX sits on the distribution side of this map while securing licensed rights deal by deal. It runs sourcing and qualification, rights review, contracting, data preparation, delivery and payment for businesses licensing operational records to AI developers. Datasets are sourced on request rather than held in stock. A supplier's records are delivered only after it has signed off on scope and terms; from then on, the signed agreement sets how SourceX may use, sublicense and distribute the de-identified dataset.

None of its operating metrics is public. SourceX is raising from accredited investors under Rule 506(c) and shares investor materials on request through its investors page, which is the place to ask the questions below. They apply to any distributor of AI data, SourceX included:

  • How many datasets does a typical buyer take through one approved relationship per year, and what does the full distribution look like?
  • What share of licensed datasets went to buyers already onboarded?
  • For each dataset sold, what was the buyer's next-best substitute, and who judged it?
  • What does opening a new buyer cost in hours, and what does adding a new supplier cost?
  • What are gross billings versus retained revenue, and what passes to rights holders?
  • What share of input cost goes to the largest rights holder, and on what notice can it leave?
  • How many fulfillment hours does each dataset take, split by buyer?

Methods note: sources, model and what was not measured

The comparison uses six public sources checked on 2026-10-27 through searches limited to each publisher's own domain: LiveRamp's annual report for the fiscal year ended March 31, 2026 and FactSet's for fiscal 2024 on sec.gov, a 2024 Copyright Clearance Center press release, a February 2026 Microsoft Advertising post, Microsoft's supplier assurance page and Cloudflare's pay per crawl announcement. Figures in the article are only those the documents state; no market size, growth rate, take rate or ranking was estimated.

The substitution test is an editorial model. Its three cases use illustrative inputs, and the classification is reproduced by a short script that compares exact fractions, saved with the article's working files. Not measured: actual values of s and t for any company, buyer onboarding costs in hours or money, and every metric for SourceX, which are not public.

Frequently asked questions

What is an AI data distribution business?

It is a company whose value comes from connecting buyers of AI data with the owners of records rather than from owning the records itself. It may run a marketplace, a collective license, a sourcing service or a gate that charges for access. Its edge is buyer access, contracting and rights evidence it can repeat across many owners.

Is owning data or distributing it the better AI data business model?

Neither in general. Ownership earns more when the records are close to unique for the buyer's task, and distribution earns more when substitutes exist and each new supplier relationship is expensive to open. The substitution test compares the owner's scarcity rent with the new-supplier cost a channel removes.

How do you compare a data marketplace that reports net revenue with a licensor that reports gross?

Convert both to retained revenue per dollar the buyer pays. LiveRamp, for example, reports its Data Marketplace revenue net because its obligation is to facilitate transactions between providers and buyers, so you need gross billings to compare it with a company that books the full price and records what owners receive as a cost.

Why do buyer onboarding costs favor distributors?

Because the cost attaches to each supplier relationship, not to each dataset. Microsoft, for instance, requires suppliers that process personal or confidential data to meet its data protection requirements with an annual self-attestation, and higher-risk processing can require independent verification. A buyer that has cleared one channel avoids repeating that work for every new owner.

Where does SourceX sit between ownership and distribution?

By its published description, SourceX runs sourcing, rights review, contracting, preparation, delivery and payment for businesses licensing records to AI developers, and sources datasets on request rather than holding stock. Each supplying company signs before delivery, and SourceX then holds the rights the agreement grants. Its operating metrics are not public; investor materials are available on request.

Sources

  • LiveRamp's annual report says it reports Data Marketplace revenue on a net basis because its performance obligation is to facilitate transactions between data providers and data buyers, and describes what its Marketplace and Other revenue includes. Source
  • FactSet's fiscal 2024 annual report says some content agreements can be terminated on one year's notice, some suppliers are competitors who may restrict use or charge prohibitive amounts, and two suppliers each exceeded 10 percent of total data costs. Source
  • Copyright Clearance Center added AI re-use rights for internal use to its enterprise-wide Annual Copyright License in 2024, as non-exclusive, annually renewable rights from participating publishers. Source
  • Microsoft Advertising's February 2026 post describes the Publisher Content Marketplace: publishers define licensing and usage terms, AI builders license content for grounding scenarios, and publishers are paid on delivered value with usage-based reporting. Source
  • Microsoft's Supplier Security and Privacy Assurance program requires suppliers that process personal or Microsoft confidential data to meet its Supplier Data Protection Requirements, with annual self-attestation and independent verification for higher-risk processing. Source
  • Cloudflare's pay per crawl lets site owners allow, charge or block AI crawlers, answers unpaid requests with HTTP 402 and pricing, and aggregates billing events, charges crawlers and distributes earnings to publishers. 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