Skip to content

AI data market

SourceX AI data company comparison: roles, rights and blank fields

By SourceX Editorial · Reviewed by Noah Loul ·

9 sections

Short answer

SourceX differs from large AI data companies mainly in what it sells and where supply comes from: it licenses records companies already hold, sourced per buyer request, and sublicenses them under its own agreements. Appen runs AI model development and evaluation projects, Scale AI builds on expert work, Reddit licenses Reddit content, Snowflake Marketplace and Hugging Face distribute others' data, and MOSTLY AI generates it. Size cannot be compared: SourceX publishes no metrics.

License terms

SourceX AI data company

This page covers sourceX AI data company comparison: roles, rights and blank fields.Illustrative

Key takeaways

  • Compare roles, not size: the seven rows sell project and expert work, platform content, distribution, generation and licensed company records, so their revenue, bookings and funding figures do not line up.
  • Appen and Reddit file audited numbers but, in the documents read, say little about the terms on which they sell; SourceX publishes how a supply deal works but no numbers.
  • SourceX's supply is records that companies already hold, sourced per buyer request; its pages say it is not a marketplace or broker and that it holds a sublicensable license once a deal closes.
  • Every blank in the evidence table is a diligence question; for SourceX the open fields are onward customer terms, deal counts, revenue and margins.
  • No ranking is possible or offered: no common metric and period exists across these companies.

How should an investor compare SourceX with large AI data companies?

Compare them by business role and by evidence level, not by size. A SourceX AI data company comparison is useful when it answers two questions for each company: what exactly is sold, and from whose supply; and what a reader can check about it, in which document, as of which date. Size, growth and customer counts cannot be compared here, because the companies report different things, some report nothing, and SourceX publishes no metrics.

That makes the decision narrower than a league table and more useful in diligence. An investor looking at SourceX wants to know whether its model is a variant of something large and familiar, a labeling workforce, a content licensor, a data marketplace, or something with different economics. The answer has to come from each company's own filings, pages, documentation and terms, read on 28 October 2026, with every field that a company does not disclose left blank rather than filled from press estimates.

Six companies that publish primary documents stand in for six roles: Appen for AI data project services, Reddit for licensing Reddit content, Snowflake Marketplace for commercial data distribution, Scale AI for expert data work, MOSTLY AI for synthetic generation and Hugging Face for open dataset hosting. None is ranked against another or against SourceX.

What each company says it sells, with the document and date

Each company describes a different product, and the descriptions below stay close to its own words. Appen's 2025 annual report, lodged with the Australian Securities Exchange on 25 February 2026, describes generative AI model development and evaluation projects for model builders, large technology and enterprise customers, and shows revenue concentrated in its top five customers. The report includes the company's financial statements; this page uses none of its figures.

Reddit's annual report on Form 10-K for 2025 describes content licensing revenue as recognized when content partners consume and benefit from their use of the licensed content, generally ratably over the license period.

Snowflake's documentation describes its Marketplace as the place where consumers explore and access listings and providers offer them, with access that can be free, a limited trial or paid, and products delivered as secure shares or applications.

Scale AI's experts page offers expert AI training data from a network of credentialed specialists. MOSTLY AI's Synthetic Data SDK page describes an open-source Python toolkit that trains a generator on an organization's own data and produces privacy-safe synthetic records, either locally on the user's compute or through MOSTLY AI's hosted platform. Hugging Face's Hub documentation says the Hub hosts community-curated datasets, each stored as a Git repository, and that the creator adds a license to the dataset card, which users are asked to respect.

SourceX's live pages describe a different kind of supply. It licenses selected records that companies already hold in their business systems, sources each dataset when a buyer asks for one rather than holding stock, and calls itself the supplier's direct licensing counterparty rather than a marketplace or broker. Its legal framework page adds that, once a deal closes, it holds a perpetual license to the de-identified dataset that it may sublicense, and that what its customers may do is set in SourceX's own agreements with them.

The dated evidence table, with blank fields left visible

The table is the asset: one row per company, the document and date behind it, the role and supply source in the company's own terms, and the fields that the documents read do not show. A blank is a finding, not a gap to fill. The SourceX row uses only its live pages.

The dated evidence table, with blank fields left visible
Company, document, date readRole in its own documentsWhere supply comes fromNot shown in the documents read
Appen (listed, ASX): 2025 annual report of 25 Feb 2026Generative AI model development and evaluation projectsProject work it delivers for model builders and other customersWho owns the outputs; terms with clients
Reddit (listed): Form 10-K for 2025Licenses Reddit content to partners; revenue recognized over the license periodReddit contentLicense terms, prices and permitted uses
Snowflake Marketplace (listed parent): documentationMarketplace where providers list data products and consumers access themThird-party providers' listingsA revenue figure for the Marketplace itself
Scale AI (private): experts pageExpert AI training dataA network of credentialed specialistsAudited financials; who owns outputs
MOSTLY AI, powered by Syntho (brand acquired by Syntho; private): Synthetic Data SDK pageToolkit that generates synthetic data from an organization's own dataThe user's own records and a trained generatorAudited financials; output terms
Hugging Face (private; NVIDIA agreed on 2 Sep 2026 to acquire it, pending when read): Hub documentation at commit f726ac8Hosting for community-curated datasetsUploaders, under licenses they chooseAudited financials; revenue from datasets
SourceX (private): live pages read 28 Oct 2026Licenses selected business records and sublicenses the de-identified datasetCompanies' own systems, sourced per buyer requestOnward customer terms, deal counts, revenue and margins

Terms or numbers: which kind of evidence each row offers

The rows differ as much in the kind of evidence they offer as in the business they describe. The grid codes three fields for each company: where its role and supply are described, whether the way its deals work is published, and whether audited numbers are filed. Filing means a securities filing; Docs means product documentation; Own page means a marketing or product page; Not found means absent from the documents read; Not public means no audited statements are published, as for any private company.

Two patterns stand out. Appen and Reddit file audited numbers, but the documents read say little about how their work or content is sold: Reddit's filing gives only revenue timing, and Appen's report gives no terms at all. SourceX inverts that: its pages publish how a supply deal works, with a Master Supplier Licensing Agreement signed once, a Dataset Deal Confirmation per deal that fixes the exact dataset, rights and processing route, a fee that is usually paid once, exclusivity for AI training and a permanent license once paid, but no numbers. Its FAQ says it does not publish its commercial terms, and its investor page offers company and commercial figures to investors privately instead of printing them.

Snowflake Marketplace and Hugging Face document the mechanics of distribution, a listing that is free, a trial or paid, or a license chosen for each dataset, while neither shows what the platform earns from it. Snowflake files audited numbers for the company as a whole; Hugging Face is private, though NVIDIA agreed on 2 September 2026 to acquire it (pending when read), and like Scale AI and MOSTLY AI it publishes none in the documents read. For an investor, the practical consequence is a different diligence list per row, not a single scorecard.

A grid of seven companies against three evidence fields, role and supply, how deals work and audited numbers: Appen, Reddit and Snowflake, the Marketplace parent, have filed audited numbers, with deal mechanics not found for Appen, partial for Reddit (revenue timing only) and partial for Snowflake; Scale AI and MOSTLY AI show neither published deal terms nor public numbers; Hugging Face and SourceX publish how deals work, and their numbers are not public.
Coding from documents read on 28 Oct 2026; Not found means absent from those documents.

Where SourceX's role differs, and where it overlaps

SourceX's role differs most in where supply comes from and in who holds the rights. Appen and Scale AI describe service businesses: generative AI model development and evaluation projects run for customers, or expert training data from a network of credentialed specialists. SourceX's pages describe licensing records that already exist inside companies, such as support conversations or sales records, so the scarce input is a company's archive and its willingness to license it, not hours of project or expert time. Work still happens on SourceX's side, qualification, rights review, contracting and de-identification through approved processing routes, and its pages describe that work as part of every deal.

Against Reddit, the difference is ownership of the source. Reddit licenses Reddit content; SourceX does not own the companies whose records it licenses, so every dataset needs a separate supplier, a separate rights review and a signed Confirmation, and its supply cannot grow by adding users to a platform. Against Snowflake Marketplace and Hugging Face, the difference is the counterparty: those platforms carry listings or uploads that others offer, with access or a license set for each listing, while SourceX takes a license itself and contracts with its customers in its own name.

The overlap is real too. Like a services firm, SourceX does labor-heavy work per dataset. Like a content licensor, it sells licenses to an asset rather than hours. Like a marketplace, it sits between holders of data and AI developers. Its own legal framework page puts it this way: SourceX is not a model developer, but it is not a neutral courier either.

Why revenue, run-rate, funding and contract value cannot fill the blanks

They measure different things, so a blank stays blank. Audited revenue counts what a company earned in a period under accounting rules; Reddit's filing shows how licensing revenue is spread over the license period, so one multi-year contract appears across several years. Run-rate figures annualize a recent period, bookings and contract value count commitments not yet earned, a marketplace's transaction value includes what providers keep, and funding measures what investors paid for shares, not what customers paid for data.

For a private company, the rule here is strict: no revenue, valuation, customer or growth figure unless the company itself published it, and even then it is reported as that company's figure for that metric and period, never lined up against another company's different metric. None of the private companies' pages read for this table publishes audited figures. Filling those cells from press estimates would turn a comparison of business roles into a comparison of rumors.

The same rule applies to SourceX with more force. Its revenue, deal count, acceptance rate and pipeline are not public, and a reader should treat any comparison that implies SourceX's scale relative to Appen, Reddit or Scale AI as unsupported. What can be compared today is the role, the rights structure and the evidence each company offers.

Risks that could make this comparison wrong

The comparison could mislead in four ways; three of them rest on a document a reader can check.

  • Self-description: every role in the table is the company's own account, read on one date. Pages change, and a company's description of itself is not proof that its business works that way.
  • Buyer concentration across the sector: Appen's 2025 annual report shows its revenue concentrated in its top five customers. If AI data spending stays with a small number of buyers, any supplier, SourceX included, depends on a few relationships.
  • Substitutes for licensed records: MOSTLY AI's toolkit generates synthetic data from an organization's own data on its own compute, and Hugging Face hosts community-curated datasets under licenses their creators choose. Where synthetic or freely available data is good enough for a task, licensed business records lose pricing power.
  • Supply that does not convert: SourceX's own pages say a company can qualify and still receive no offer, that positive commercial interest is not yet a purchase, and data that does not reach Closing earns no payment. How often each step converts is not public.

Questions to put to SourceX and to any AI data company

Ask each company the questions its blank cells raise, and ask SourceX the ones its published terms make answerable. SourceX's raise runs under Rule 506(c), a route that limits purchasers to verified accredited investors, and its materials go out on request through sourcex.si/investors; the questions below are the ones those materials should answer.

  • Which document defines what the customer may do with the data or work, and may an investor read a redacted example?
  • Is the AI data activity reported as its own line, and over what period is its revenue recognized?
  • Who owns what is produced or licensed, the company, its customer or the original source?
  • For SourceX: how many Dataset Deal Confirmations were signed and how many reached Closing, counted over the same period?
  • For SourceX: how many times has each closed dataset been licensed onward, and what share has never been licensed onward?
  • For SourceX: what share of revenue comes from its largest customers, and how does that compare with the concentration Appen reports for its own revenue?
  • For every company: which figures in its pitch are audited, which are bookings or run-rate, and which are estimates?

Methods note: documents read, coding rules and what was not measured

Documents: Appen's 2025 annual report (ASX announcement of 25 February 2026) and Reddit's and Snowflake's annual reports on Form 10-K, verified through search excerpts from announcements.asx.com.au and sec.gov because direct fetches were blocked; Snowflake's Marketplace documentation, Scale AI's experts page and MOSTLY AI's SDK page, verified the same way on their own domains, as were Syntho's release on the MOSTLY AI brand and NVIDIA's Form 8-K on Hugging Face, the sources of the two status notes in the table; Hugging Face's Hub documentation read from its official repository at commit f726ac870c1c; SourceX's live pages fetched on 28 October 2026. All dates of reading are 28 October 2026.

Coding rules: a role or supply source is recorded only when the company's own document states it; a field is Not found when none of the documents read states it, which does not prove the company has never disclosed it; Not public marks a private company that publishes no audited statements. How deals work is coded Docs or Own page when the document describes the terms or mechanics of a sale or license, such as the license attached to a dataset or SourceX's typical deal terms, and partial when it describes only one element, such as revenue timing for Reddit or the free, trial or paid access of a Snowflake listing. The selection of six companies is editorial: one well-documented company per role, chosen from companies that publish primary documents, not a sample of the market.

Not measured: revenue, growth, customers, headcount, valuation and market share, for any company; the quality of any company's data; whether any company's description matches its practice. No figure from any company's financial statements is used.

Frequently asked questions

How is SourceX different from Scale AI or Appen?

Scale AI describes expert AI training data from a network of credentialed specialists, and Appen's annual report describes model development and evaluation projects for model builders and other large customers. SourceX's pages describe licensing records that companies already hold and sublicensing the de-identified dataset under its own agreements. Both involve labor, but the scarce input differs: project and expert hours on one side, a company's existing archive and its consent to license it on the other.

Is SourceX a data marketplace like Snowflake Marketplace or Hugging Face?

Its own FAQ says it is not a marketplace or broker. Snowflake Marketplace and Hugging Face carry listings or uploads that others offer, with access or a license set for each one, while SourceX takes a license from the supplying company and contracts with its customers in its own name.

Can SourceX's size be compared with these companies?

No. SourceX publishes no revenue, deal counts or growth figures; its investor page keeps company and commercial figures for investors who request materials. Any comparison implying its scale relative to listed or private peers is unsupported.

Why not fill private companies' blanks with press estimates?

Press estimates mix revenue, run-rate, bookings and funding, which measure different things over different periods, and they cannot be checked against the company's own documents. This comparison records a figure only when the company published it, and it uses none here.

What should an investor ask SourceX first?

Ask how many Dataset Deal Confirmations reached Closing over a stated period, how many times each closed dataset has been licensed onward, and how concentrated revenue is among its largest customers. Those numbers are not public and decide whether the licensing model behaves differently from a services business.

Sources

  • Appen's 2025 annual report (ASX, 25 February 2026) describes generative AI model development and evaluation projects for model builders, large technology and enterprise customers, and shows revenue concentrated in its top five customers. Source
  • Reddit's 2025 Form 10-K describes content licensing revenue recognized generally ratably over the license period. Source
  • Snowflake, the Marketplace's parent, files an annual report on Form 10-K, here for the fiscal year ended 31 January 2026, with audited financial statements. Source
  • Snowflake Marketplace documentation: providers offer listings to consumers, access can be free, a limited trial or paid, and products are delivered as shares or applications. Source
  • Scale AI offers expert AI training data from a network of credentialed specialists. Source
  • MOSTLY AI's Synthetic Data SDK is an open-source toolkit that generates synthetic data from an organization's own data, locally or through its hosted platform. Source
  • Syntho's website carries a release headlined Syntho Acquires MOSTLY AI Brand, saying the brand continues as MOSTLY AI, powered by Syntho; its date could not be confirmed. Source
  • Hugging Face Hub documentation: community-curated datasets stored as Git repositories, with a license added by the creator in the dataset card. Source
  • NVIDIA's Form 8-K reports that on September 2, 2026 NVIDIA entered into a definitive agreement to acquire Hugging Face, Inc., with closing subject to customary conditions including required regulatory approvals. Source
  • SourceX says it is not a marketplace or broker but the supplier's direct licensing counterparty, and that it does not publish its commercial terms. Source
  • SourceX holds a perpetual, sublicensable license to the de-identified dataset after closing, and its customers' rights are set in its own agreements with them. Source
  • SourceX's typical deal terms: one Master agreement, a Dataset Deal Confirmation per deal, usually a one-time fee, exclusivity for AI training and a permanent license once paid; a company can qualify and still receive no offer. Source
  • SourceX provides company and commercial metrics to investors on request instead of publishing them. Source
  • SourceX sources datasets on request rather than holding them in stock. 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