AI data market
Where SourceX fits in the AI data value chain, read from its own terms
By SourceX Editorial · Reviewed by Noah Loul ·
9 sections
Short answer
SourceX sits at the licensing link of the AI data value chain. By its own pages, it licenses selected business records directly from companies, runs qualification, rights review, contracting and preparation, and licenses the de-identified dataset onward to AI developers in its own name. It sources datasets on request rather than from stock, fixes a processing route in each deal, and publishes no commercial results, so its position has to be tested in diligence.
License terms
SourceX AI data value chain
Key takeaways
- SourceX is a principal on both sides of the licensing step: it licenses in from companies and onward to AI developers in its own name, and says it is not a marketplace or broker.
- Its published process spans four of six links, from qualification to onward licensing; record creation belongs to the supplier and labeling, training and evaluation to the buyer.
- Records move only after SourceX signs its own onward agreement, and de-identification runs through SourceX or its processing partner on a route fixed per deal.
- Neighboring links are separate businesses: annotation platforms split tooling, labor and rights, and EU AI Act Article 25 formalizes written handoffs for high-risk systems.
- No SourceX findings or metrics are public; each link has a fixed-denominator metric an investor can ask for instead.
Where does SourceX fit in the AI data value chain?
SourceX sits at the licensing link: between the company whose systems hold the records and the developer who trains or evaluates a model on them. Its investor page, read on 27 October 2026, says SourceX licenses business data directly from companies and licenses it onward to AI developers in its own name, managing sourcing, rights review, contracting, preparation, delivery and payment.
The phrase in its own name carries most of the weight. SourceX's FAQ says it is not a marketplace or broker that finds buyers for a company's data, but the supplier's direct licensing counterparty from assessment through payment. In value chain terms that makes it a principal on both sides of the licensing step, not a venue that matches two parties and steps aside.
An investor's decision is whether that position can hold durable value, and the public record only goes so far. This article maps the chain link by link, records what SourceX's pages say and when they were read, describes the neighboring links from their own documents, and lists the evidence a diligence process would need. No SourceX commercial figures appear because none are published.
Six links from business system to trained model, and SourceX's span
SourceX's published process covers links two to five of a six-link chain; the first belongs to the company that created the records and the sixth to the buyer and its tools. The table assigns rights, work and the governing document at each link from SourceX's own pages, and gives each link a metric whose denominator is fixed in advance, so it can be compared with any other company's.
Two details define the span. On the supply side, the how-it-works page describes one Master Supplier Licensing Agreement signed once after qualification and a Dataset Deal Confirmation per deal that fixes the exact dataset, rights, processing route and deadlines; returning suppliers reuse the Master, and the supplier signs off on scope and terms before anything is delivered. On the buyer side, SourceX seeks commitment on the approved terms and signs its own onward agreement before it instructs processing, which means records move only once a buyer has committed.
Preparation is the link where SourceX shares the work. De-identification is performed by SourceX directly or through its processing partner, along the route the Confirmation fixes, and the FAQ adds that the supplier agreement records that re-identification may still be possible. After the deal, the buyers page says SourceX handles future purchases and the supplier relationship.
| Link | Who holds the rights | Who does the work | Governing document | Metric with a fixed denominator |
|---|---|---|---|---|
| 1. Record creation | Source company, which keeps ownership of its originals | Source company's staff and systems | Its own customer, vendor and privacy terms | Share of inventoried systems that clear rights review |
| 2. Qualification and inventory | Unchanged; nothing is licensed yet | SourceX with the company; no records shared | Application and fit check, no obligation | Qualified companies per completed assessment |
| 3. Rights review and contracting | License to SourceX for the selected data; AI training exclusivity binds the supplier | SourceX, with the supplier's counsel where needed | Master once; Dataset Deal Confirmation per deal | Signed Confirmations per signed Master |
| 4. Preparation and de-identification | SourceX owns the de-identified dataset, subject to the supplier's IP | SourceX or its processing partner | Processing route fixed in the Confirmation | Closed datasets per signed Confirmation |
| 5. Onward licensing and delivery | SourceX licenses in its own name and may use, sublicense and distribute as agreed | SourceX | SourceX's onward agreement with the buyer | Distinct buyers per closed dataset |
| 6. Labeling, training and evaluation | The buyer, within its license | Buyer, its tools and its workforces | Buyer's programs; EU AI Act duties for high-risk providers | Not SourceX's to report |
What SourceX's own pages say, with the date they were read
Every statement about SourceX in this article comes from a page fetched on 27 October 2026 or from SourceX's approved public facts. The right-hand column matters as much as the middle one: it is the evidence a page cannot supply, which is where diligence starts.
One line deserves emphasis. The research methodology page says SourceX's own survey and transaction research programs are in development and that no SourceX findings have been published. Any claim about SourceX's volumes, conversion or buyers therefore has no public basis today, and this article makes none.
| What the page says | Page, read 27 Oct 2026 | What it does not show |
|---|---|---|
| Licenses data directly from companies and onward to AI developers in its own name | Investors | Which developers, volumes or prices |
| Not a marketplace or broker that finds buyers for a company's data | FAQ | How often onward demand exists before a supplier signs |
| Signs its own onward agreement, then instructs processing | How it works | Onward terms, margins or buyer count |
| Usually a one-time fee to the supplier; some deals set recurring deliveries; no royalties | How it works | Share of deals with recurring deliveries |
| Owns the de-identified dataset subject to supplier IP; license permanent once paid | How it works and FAQ | How many times a dataset is licensed onward |
| Sources datasets on request, not from stock; request form lists frontier model developers, AI data labs and training or evaluation partners | Request data | Which organization types actually buy |
| No SourceX research findings published yet | Research methodology | Survey or transaction results |
The neighboring links, described by their own documents
The links on either side of SourceX are run by different kinds of firms, and their own documents show where the boundaries fall. NIST's AI Risk Management Framework places data work among AI Design tasks performed during the application context and data and input phases, including gathering and cleaning data and documenting dataset metadata, with data engineers and data providers among the named actors. SourceX's link is the provider role in that list, extended to include the license itself.
Annotation sits downstream, on the buyer's side. Amazon's SageMaker Ground Truth documentation describes a workforce as the group of workers chosen to label a dataset, drawn from Amazon Mechanical Turk, a vendor-managed workforce or the customer's own private workforce, with SageMaker sending the tasks; the same page announced that Mechanical Turk would permanently close on September 30, 2026. Tooling, labor and data rights are three separate suppliers there, and none of them is the company that created the records.
Regulation is starting to formalize the handoffs. Article 25 of the EU AI Act, titled responsibilities along the AI value chain, requires the provider of a high-risk AI system and any third party supplying tools, services, components or processes used in it to set out by written agreement the information, capabilities, technical access and other assistance the provider needs, with an exception for tools released under free and open-source licenses other than general-purpose AI models. The duty applies to high-risk systems only, but it points to documentation flowing upstream to whoever supplied the inputs.
What the licensing link adds, and how to test each claim
Read from its own pages, SourceX's link adds three things the neighboring links do not: a single counterparty for each side, a contract form that repeats, and anonymity for the buyer. Each is a claim an investor can test rather than a result anyone has published.
A single counterparty: the supplier's contract counterparty is SourceX from assessment to payment, and the buyer receives a license in SourceX's name, so the two sides never need to negotiate with each other. The test is whether supplier Confirmations and onward agreements line up closely enough that a dataset licensed in from one company can be licensed out without reopening its rights.
A repeating contract: one Master per supplier and a short Confirmation per deal, which the investor page presents as one repeatable agreement that shortens legal review. The test is legal time per Confirmation across successive deals, tracked separately for new and returning suppliers.
Buyer anonymity: the investor page says labs can source data without revealing their roadmap or identity to the market. In a chain where a data request can signal a model's next capability, that is a genuine function. The test is whether suppliers still sign when they cannot know the end buyer, measured as signed Confirmations per qualified supplier.
Principal, platform or services firm? Testing the label
SourceX describes itself as a direct licensing counterparty, and the strongest alternative reading is that much of the value it adds is skilled services. Its pages describe guided exports, structured rights questions, inventory work with each company and preparation through processing routes, all of which take people. Its investor page counters with two repeatability claims: one repeatable agreement meant to shorten legal review on every deal, and ownership checks and removal of personal details built into the workflow rather than handled ad hoc.
Neither reading can be settled from public pages, but the evidence that would settle it is easy to name. Labor hours per closed dataset should fall as deals accumulate if the standardized contract and tooling work. Returning suppliers reuse the Master, so the share of Confirmations signed under an existing Master shows whether supply repeats. The share of preparation handled by the processing partner rather than SourceX's own pipeline shows how much of the cost sits outside the company. An investor should ask for each as a time series with its denominator.
Risks that could weaken SourceX's position in the chain
Each risk below sits at a specific link, and several come straight from SourceX's own terms. A company that cannot say which one would hurt it most has not tested its position:
- Supply that does not close. Data that does not reach Closing earns no payment, and declined data is deleted, so every deal carries a chance of ending after work has been done on both sides.
- Each sale needs new supply. A usually one-time supplier fee and a permanent license once paid mean growth depends on sourcing new datasets, not on renewals of old ones.
- Residual re-identification risk. The FAQ states that no de-identification method is perfect and that the supplier agreement records re-identification may still be possible; buyers and regulators may push that risk back toward the licensor.
- Processing dependency. De-identification runs through SourceX directly or its processing partner, so partner capacity and quality become part of SourceX's own execution risk.
- Upstream documentation duties. Article 25 of the EU AI Act, for high-risk systems, makes written information duties part of supplying inputs, which could raise the documentation cost of every onward license.
- Neighbors moving into the link. Annotation platforms already separate tooling, labor and rights; a firm that controls labor or tooling could add licensing, and buyers with private workforces can prepare data themselves.
Diligence questions about the licensing link
Ask for evidence at the links where SourceX claims its role, using definitions you set rather than ones the company chooses. The list below follows the table's metrics, so the answers can be compared with any other licensor's.
SourceX's present raise is open only to verified accredited investors under Rule 506(c) of Regulation D, and materials are requested through sourcex.si/investors. The questions an investor can bring to that conversation:
- Of companies that completed an assessment, how many qualified, and how many of those signed a Master?
- Of signed Confirmations, how many closed, how many ended in a no-deal notice, and what were the stated reasons?
- How many closed datasets went to more than one buyer, and how is each onward license scoped?
- What share of Confirmations came from returning suppliers under an existing Master?
- Which processing route handled each closed dataset, and what did preparation cost per dataset on each route?
- Which documentation does a buyer receive per dataset, and how would it change for a high-risk system under the EU AI Act?
Methods note: pages read, sources and what was not measured
SourceX statements come from its investors, how-it-works, FAQ, buyers, request-data and research methodology pages, fetched directly on 27 October 2026, and from the approved public facts register. Descriptions of other links come from NIST's AI RMF Appendix A, Amazon's SageMaker Ground Truth workforce documentation and the EU AI Act on EUR-Lex. The six-link chain, the table's role assignments and its metrics are an editorial framework built for this article on the same date.
Not measured: SourceX's volumes, conversion at any link, buyer counts, prices, margins or labor per dataset, and the same figures for any other company. None are public for SourceX, and the methodology page states that no SourceX findings have been published. Competitors are described only through their own primary documents, with no figures attributed to them.
Frequently asked questions
Is SourceX a data marketplace?
By its own description, no. SourceX's FAQ says it is not a marketplace or broker that finds buyers for a company's data, but the company's direct licensing counterparty, and its investor page says it licenses datasets onward in its own name.
Who de-identifies the data in a SourceX deal?
SourceX does it directly or through its processing partner, along the processing route fixed in each Dataset Deal Confirmation. The FAQ notes that no method is perfect and that the supplier agreement records that re-identification may still be possible.
Does SourceX hold datasets in inventory?
No. Its request-data page says datasets are sourced on request, not held in stock, and its buyers page describes example demand categories rather than inventory under contract.
Which organizations does SourceX license data to?
SourceX licenses datasets onward to AI developers in its own name. Its data request form lists frontier model developers, AI data labs and training or evaluation partners as organization types; it names no buyers publicly, and this article names none.
Which SourceX metrics are public?
None of its commercial or operating metrics. The research methodology page says no SourceX findings have been published, and the investor page says company and commercial metrics are shared with investors directly.
Sources
- SourceX's investor page says it licenses business data directly from companies and licenses it onward to AI developers in its own name, uses one repeatable agreement and shares metrics with investors directly. Source
- SourceX's how-it-works page describes the Master and Dataset Deal Confirmation, the onward agreement before processing, the usual one-time fee, ownership of the de-identified dataset, exclusivity and the permanent license. Source
- SourceX's FAQ says it is not a marketplace or broker but a direct licensing counterparty, and that the supplier agreement records re-identification may still be possible. Source
- SourceX's request-data page says it licenses datasets in its own name, sources them on request rather than from stock, and lists organization types on its form. Source
- SourceX's buyers page says it handles future purchases and the supplier relationship and describes example demand rather than inventory under contract. Source
- SourceX's research methodology page says its survey and transaction research programs are in development and no SourceX findings have been published. Source
- NIST AI RMF Appendix A places AI Design tasks, including gathering and cleaning data and documenting dataset metadata, in the application context and data and input phases, with data engineers and data providers among the actors. Source
- Amazon's SageMaker Ground Truth documentation describes Mechanical Turk, vendor-managed and private workforces, with SageMaker sending tasks, and announced Mechanical Turk's closure on September 30, 2026. Source
- Article 25 of the EU AI Act requires providers of high-risk AI systems and third-party suppliers of tools, services, components or processes to agree in writing on the information and assistance the provider needs. Source
Request SourceX investor materials
SourceX shares company and commercial metrics with investors directly. Any offering is made only to verified accredited investors.