Skip to content

AI data market

448 suppliers, no market: data aggregation marketplace economics

By SourceX Editorial · Reviewed by Noah Loul ·

8 sections

Short answer

Aggregating data suppliers does not by itself create marketplace economics. A marketplace exists when the platform matches buyers with competing suppliers, sets terms both sides accept, and each side gains as the other grows. The Registry of Open Data on AWS gathers 1,244 listings from 448 managers yet records no price, buyer or transaction. Test matching, not supplier counts, before accepting a marketplace label.

Value drivers

448 suppliers, no market

This page covers 448 suppliers, no market: data aggregation marketplace economics.Illustrative

Key takeaways

  • A supplier count measures supply; marketplace economics depend on matching, which a supplier count cannot show.
  • In the Registry of Open Data on AWS, one manager supplies 24.6% of 1,244 listings and 80.4% of managers list a single dataset.
  • Demand evidence concentrates too: the top tenth of listings carries 43.0% of the registry's recorded usage entries.
  • Revenue booked net of a supplier's share points to a facilitator; gross revenue with sourcing staff in cost of revenue points to a reseller or services firm.
  • Exclusive licensing caps a dataset at one buyer, so a principal licensor should be judged on transaction economics, not network effects.

Is a data aggregator a marketplace once it has enough suppliers?

No. A supplier count describes supply, while marketplace economics come from matching: buyers arrive with needs, more than one supplier can meet them, the platform sets terms both sides accept, and growth on one side makes the platform more useful to the other. An aggregator missing those features is a catalog, a reseller or a service desk, and each of those earns different margins and carries different risks.

Economics offers a crisp version of the test. In their 2006 RAND Journal of Economics paper, Jean-Charles Rochet and Jean Tirole define a two-sided market as one where the volume of transactions between end users depends on how the platform splits its charges between the sides, not only on the total it charges. If moving charges from suppliers to buyers would leave trade volume unchanged, the platform is not running a two-sided market, whatever its listing count.

Regulators look at the same thing from another angle. The EU's platform-to-business regulation, 2019/1150, covers services that let business users offer goods or services to consumers in order to facilitate direct transactions between them, provided under contracts with those business users. Pure business-to-business services fall outside it, so it does not govern data deals between companies, but its definition puts the weight on who transacts with whom rather than on how many sellers appear in a directory.

What 1,244 open listings from 448 managers reveal

The Registry of Open Data on AWS shows aggregation at scale with no market attached. At commit 273de0e (27 October 2026) its repository holds 1,244 dataset listings from 448 managing organizations, and the listing schema has fields for a name, description, license, contact, resources and usage examples but none for a price, a buyer or a transaction.

Supply is concentrated. NASA manages 306 listings (24.6%), NOAA 81 and AODN 58. The ten largest managers hold 576 listings (46.3%), while 360 managers (80.4% of all managers) list exactly one dataset. A headline of 448 suppliers therefore describes a few anchor contributors and a long tail.

Recorded demand concentrates as well. Each listing can carry DataAtWork entries, which the README describes as tutorials, tools and applications, and publications that use the data. Of 3,404 such entries, the 125 listings with the most entries, the top tenth, hold 1,463 (43.0%), and the median listing has two. In all, 375 listings (30.1%) have no entry, and most of them are recent: among 882 listings added before 2026, 141 (16.0%) have none, while among 362 added in 2026, 234 (64.6%) have none. NASA alone added 229 of those 362. Entries show where someone documented a use, not who paid, so treat them as a proxy.

The economics sit outside the listings. AWS's Open Data Sponsorship Program page says the program covers the cost of storing and sharing publicly available, high-value, cloud-optimized datasets on the registry, and the README notes that the datasets are not provided by AWS while pointing users to AWS compute and analytics products. This is an inference, not a disclosed figure: any return to AWS arrives through use of its cloud, not a cut of data trades. For 326 listings (26.2%), access to at least one resource runs through a controlled-access process set by the data provider, outside the registry.

Left, a treemap of 1,244 registry listings by manager: NASA 306, NOAA 81, AODN 58, seven more top-ten managers 131 and 438 other managers 668. Right, a concentration curve of 3,404 usage entries: the top tenth of listings holds 43.0 percent, and the curve runs flat across 375 listings with no entries.
Registry of Open Data on AWS at commit 273de0e, 27 October 2026; counts by SourceX from public repository data. The listing schema has no price, buyer or transaction field.

Four places an aggregator can keep the value

An aggregator keeps value in one of four places, and only one of them is a take on matched trades. Sort a company by who holds the rights at the moment of sale, who does the work on each order, which tooling it supplies and where recurring revenue lands, then ask for the metric that fits the model it actually runs.

A sponsor aggregator, like the AWS registry, subsidizes supply to sell something else. A revenue-share marketplace takes a cut of trades it facilitates: LiveRamp's annual report for the fiscal year ended March 31, 2026 says its Data Marketplace revenue comes primarily through revenue-sharing arrangements with data sellers and is recognized net of the share earned by the seller. A principal licensor buys or licenses rights and sells in its own name, which is how SourceX describes itself on its investor page, and it usually pays a one-time fee per dataset, with some agreements setting recurring deliveries. A services desk calls itself a marketplace but has its own staff source, clean or label each order.

The table normalizes each model to the unit an investor should ask about. Public sources describe roles, not volumes, so the right-hand column says what the record shows and what still has to come from the company.

Four places an aggregator can keep the value
ModelRights holder and contract partyWork per orderNormalized metric to requestPublic record checked
Sponsor aggregator (AWS registry)Each data provider; the registry contracts with no buyerProvider maintains data; host covers storage and sharingUsage entries per listing and anchor-supplier share1,244 listings, 448 managers, no price field (count above)
Revenue-share marketplace (LiveRamp Data Marketplace)Data seller; platform facilitates the tradeSeller supplies the data; platform takes a revenue shareNet revenue per matched trade and offers per buyer requestRevenue share recognized net of the seller's portion (FY2026 10-K)
Principal licensor (SourceX, by its own pages)Supplier licenses to the licensor, which licenses onward in its own nameLicensor runs sourcing, rights review, contracting and preparationGross margin and labor hours per licensed datasetExclusive AI training licenses; no metrics published
Services desk labeled a marketplace (illustrative)Varies by order; often the buyer commissions new workStaff collect, clean or label for each requestStaff hours and cost of revenue per orderNot tied to a named company; a pattern to test

The matching test: five checks before the label

Ask for five measurements that only a working marketplace can produce, each over a stated period and denominator. A company that cannot produce them may still be a good business, but it should be valued as the model it runs rather than the one it names in its deck.

Run the checks in order; the first failure usually tells you which of the four models you are looking at:

  • Choice: the share of buyer requests that drew two or more qualifying supplier offers. One offer per request means the platform routes rather than matches.
  • Reach: distinct buyers per supplier asset per year. A count stuck at one, by design or by exclusivity, removes the reuse that makes supplier growth valuable to buyers.
  • Contracting locus: the share of transaction value signed on platform terms, and whether the platform or the supplier is the counterparty. Controlled access arranged off-platform, as in a quarter of the registry, fails this check.
  • Price structure: evidence from past changes in supplier or buyer charges and the volume that followed. If volume tracked only the total charge, the Rochet-Tirole test says the market is one-sided.
  • Cross-side benefit: fill rate or time to fill for buyer requests, split by the size of the supplier pool available at the time. Flat results as the pool grows mean more suppliers added cost, not value.

Is the marketplace label covering a services business?

Often it is, and revenue accounting is the first place it shows. A facilitator reports only its share of each trade, which is how LiveRamp describes its data marketplace revenue; a company that books the full order value and carries sourcing, cleaning or labeling staff in cost of revenue is acting as a reseller or a service firm, whatever its pitch calls it.

Three follow-ups settle most cases. Ask for revenue recognized gross versus net by product line, for cost of revenue split into supplier payments, labor and tooling, and for the share of orders that required new collection or cleaning rather than delivery of an existing asset. A rising labor share as volume grows is the signature of a services desk.

None of this makes a services model a bad investment. It changes the comparable set, the margin ceiling and the scaling question, from how fast matching improves to how fast skilled work can be hired, priced and kept busy.

Running the matching test on SourceX

SourceX does not present itself as a marketplace, and its published terms explain why most of the matching test does not apply. Its investor page says it licenses business data directly from companies and licenses it onward to AI developers in its own name, and that labs can source data without revealing their roadmap or identity to the market. Buyers and suppliers do not meet on a shared listing.

The reach check is capped by design. Its how-it-works page says licenses are exclusive for AI training, so a supplier agrees not to license the selected data to anyone else for that purpose, and each deal runs on a Dataset Deal Confirmation under one Master Supplier Licensing Agreement. Choice and fill happen inside the company, which runs sourcing, qualification, rights review, contracting, preparation, delivery and payment.

That places SourceX in the principal-licensor row, and the investor question becomes transaction economics: margin per licensed dataset, labor hours per deal, how often suppliers return under the same Master, and how many distinct buyers place requests. SourceX publishes none of these figures and keeps its company metrics for conversations with investors. Its current raise is open only to accredited investors under Rule 506(c), and the investors page takes requests for materials.

What would make this reading wrong

Four developments would weaken the case that aggregation alone earns no marketplace premium, or would change how it applies. Each can be tracked.

  • Free sponsored supply could undercut paid aggregation of commodity data. AWS covers storage and sharing costs for qualifying open datasets, so an aggregator of similar data competes with a zero price, and mainly rights-restricted or private records escape that comparison.
  • Anchor suppliers can leave. When one manager accounts for a quarter of listings, as NASA does in the registry, a single departure reshapes the catalog; ask any aggregator for its top-supplier share and contract terms.
  • Matching could emerge later. If licensing terms and data descriptions become standard enough that several suppliers can answer the same request, choice and reach would rise and the five checks would start to pass; rerun them each year rather than settling the label once.
  • The proxy may mislead. Usage entries undercount real use, and open data behaves differently from licensed private records, so the registry illustrates the gap between supply and matching rather than measuring it for paid markets.

Questions for any aggregator, and how the registry count was made

Bring the five checks to every company that calls itself a data marketplace, SourceX included, and ask for the figures behind them with the period and denominator stated. The questions below are the short form.

Method: bin/registry_count.py read only the YAML files in datasets/ of a clone of awslabs/open-data-registry at commit 273de0e, dated 27 October 2026. Every file counts as one listing; 26 carry a deprecation flag or notice, and excluding them gives 47.1% for the ten largest managers and 42.8% for the top tenth's share of usage entries. A manager is the ManagedBy text, using the words inside the first pair of square brackets when it holds a link, with spaces collapsed and case ignored. Usage entries are the items under DataAtWork tutorials, tools and applications, and publications. The top tenth is the 125 listings with most entries, ten percent of 1,244 rounded up. A listing counts as controlled access or requester pays when any of its resources has that attribute. Year added is the year of the earliest commit, over full history with rename detection, that created the file or the file it was renamed from. Percentages are rounded once, half up.

Not measured: prices, transactions, buyer counts or revenue on any platform, including AWS Data Exchange, LiveRamp and SourceX, none of which the registry records. The four-model table and the five checks are editorial frameworks; the services-desk row is an illustrative pattern, not a company.

  • What share of buyer requests in the last four quarters drew two or more qualifying supplier offers?
  • How many distinct buyers did your median supplier asset reach last year, and how many reached exactly one?
  • What share of transaction value was signed on your paper, and in how many deals were you the counterparty?
  • What happened to volume the last time you changed supplier or buyer charges?
  • How do fill rate and time to fill change with the number of suppliers available for a request?
  • What share of revenue is recognized gross, and what share of cost of revenue is your own labor?

Frequently asked questions

What is the difference between a data aggregator and a data marketplace?

An aggregator collects listings or datasets from many suppliers; a marketplace matches buyers with competing suppliers under terms it sets, and each side benefits as the other grows. An aggregator can grow large without matching anyone: the Registry of Open Data on AWS lists 1,244 datasets from 448 managers but records no price, buyer or transaction.

Do data marketplaces have network effects?

Only when added suppliers improve outcomes for buyers and added buyers improve outcomes for suppliers. Supplier counts alone do not show that. Ask for offers per request, buyers per supplier asset and fill rate by supplier-pool size before crediting a network effect.

How should an investor read a data marketplace's revenue?

Check whether revenue is booked net of the supplier's share, as a facilitator would, or at the full order value with sourcing staff in cost of revenue, as a reseller or services firm would. LiveRamp, for example, says it recognizes data marketplace revenue net of the data seller's share.

Is an exclusive data licensor a marketplace?

Not in the matching sense. Exclusivity caps each dataset at one licensee, so supplier growth does not widen choice for buyers of that dataset. Judge such a company on margin and labor per licensed dataset, supplier return rates and the number of distinct buyers.

Why use an open data registry to study marketplace economics?

It is a large aggregator whose full listing data is public and versioned, which lets anyone reproduce supply and usage counts. Its lesson is about the gap between aggregation and matching; it does not measure prices or transactions in paid data markets.

Sources

  • The Registry of Open Data on AWS repository at commit 273de0e: 1,244 listing files from 448 managers, the listing schema fields, manager and usage-entry concentration, year added and access attributes, as counted for this article. Source
  • The registry README says the datasets are not provided by AWS but owned and maintained by others, points to AWS compute and analytics products, and defines DataAtWork entries as tutorials, tools and applications, and publications. Source
  • AWS says its Open Data Sponsorship Program covers the cost of storing and sharing publicly available, high-value, cloud-optimized datasets on the Registry of Open Data on AWS. Source
  • LiveRamp's 10-K for the fiscal year ended March 31, 2026 says Data Marketplace revenue comes primarily through revenue-sharing arrangements with data sellers and is recognized net of the share earned by the data seller. Source
  • Rochet and Tirole, Two-Sided Markets: A Progress Report, define a two-sided market as one in which transaction volume depends on the structure and not only the overall level of the platform's fees. Source
  • Regulation (EU) 2019/1150 defines online intermediation services as services that let business users offer goods or services to consumers to facilitate direct transactions, under contracts with business users; pure B2B services are outside its scope. Source
  • SourceX says it licenses business data directly from companies and licenses it onward to AI developers in its own name, that labs can source data without revealing their roadmap or identity to the market, and that it shares company and commercial metrics with investors directly. Source
  • SourceX's published terms: one Master Supplier Licensing Agreement and a Dataset Deal Confirmation per deal, licenses exclusive for AI training, and usually a one-time fee per dataset with some agreements setting recurring deliveries. 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