Skip to content

AI data market

Portable data competitive advantage: which of seven moat layers survive

By SourceX Editorial · Reviewed by Noah Loul ·

10 sections

Short answer

Portable data competitive advantage rarely lives in the records themselves. GDPR Article 20, the EU Data Act and the CFPB's 2024 open-banking rule let customers move records they provided or generated, but leave out derived data and cross-customer views, and two of them restrict what recipients may do. Moats that survive sit in those gaps: aggregation, outcome labels, licensing authority, accepted deliveries and repeat buyers.

Value drivers

Portable data competitive

This page covers portable data competitive advantage: which of seven moat layers survive.Illustrative

Key takeaways

  • Portability rules move the records a person or user provided or generated; derived data and cross-customer views stay with the holder.
  • A layer counts as a moat only if it does not move and survives a test designed to break it.
  • Licensing authority usually outlasts portability, because recipients of ported data face use limits and authority cannot be granted retroactively.
  • Aggregation protects only when it is both licensable and spread across many customers.
  • Concentrated buyers and changing rules, including the CFPB's 2025 reconsideration, are the clearest ways the thesis could fail.

Which data advantages survive when customers can take their records?

The advantages that survive are the ones a ported copy cannot carry: coverage across many customers, labels and outcomes the holder created, documented authority to license, a record of deliveries buyers accepted, and buyers who come back. The raw records a customer provided or generated are the weakest layer, because portability rules and contract exit clauses are written to move exactly that.

The investor decision is narrow. Portability does not ask whether a data company holds valuable records; it asks whether the value would remain if every customer exercised every export right tomorrow. A useful answer needs four kinds of evidence: what a rival would spend to rebuild each layer, what the documents let the holder license, whether the surviving layers beat raw records on a buyer's own test, and whether buyers renew.

What follows sets out what current portability rules move and what they leave behind, a seven-layer survival scorecard with a test that can break each layer, the competing explanation that useful data is still cheap to substitute, and the questions to put to any data company, SourceX included.

What portability rules move, and what they leave out

Every major portability regime moves records tied to one person or one user, and none of them moves the holder's derived work or its view across customers. Read each rule for its scope limits and for the conditions it puts on whoever receives the data; both decide what a ported copy is worth to a rival.

GDPR Article 20 lets a data subject receive the personal data they provided to a controller in a structured, commonly used and machine-readable format, and transmit it to another controller, directly where technically feasible. The Article 29 Working Party's portability guidance (WP242) reads provided data broadly enough to include data generated by a person's activity, but excludes data derived or inferred from it, using a profile built from raw smart-meter readings as its example. The European Commission's 2022 impact assessment for the Data Act judged that use of the GDPR right had proven largely theoretical, and noted that it reaches only personal data processed on the basis of consent or a contract.

The EU Data Act, applicable since 12 September 2025, gives users of connected products free access to the data their use generates and lets them have it shared with a third party of their choice. The Commission's explainer places inferred and derived data outside its scope and says data obtained this way cannot be used to develop a competing connected product. In the United States, the CFPB's October 2024 personal financial data rights rule lets consumers authorize third parties to access their financial data, but limits those third parties to collection, use and retention reasonably necessary for the product the consumer asked for, and treats targeted advertising, cross-selling and selling the data as outside that standard. The CFPB opened a reconsideration of the rule in August 2025, so its final shape is still moving.

The table compares four regimes on the questions that matter for a moat:

What portability rules move, and what they leave out
RegimeWhat a user can moveWhat the rule leaves outConditions on the recipient
GDPR Article 20Personal data the person provided, including activity data, in a machine-readable formatDerived and inferred data, such as profiles built from raw readings (WP242)Applies only to processing based on consent or a contract, per the Commission's 2022 assessment
EU Data Act, applicable since 12 September 2025Raw and pre-processed data generated by use of a connected product or related service, free of chargeInferred and derived dataNo use to develop a competing connected product; trade secrets withheld only where serious economic damage is highly likely
CFPB personal financial data rights rule, October 2024, under reconsideration since August 2025Covered financial account data, shared with third parties the consumer authorizesUses the consumer did not request, unless separately authorized as a standalone productUse limited to what the requested product needs; targeted advertising, cross-selling and data sales fall outside; authorization lasts up to one year
OMB memorandum M-25-22 for federal AI purchases, April 2025Not a user right: agencies are told to seek data and model portability and clear licensing in AI contractsWhatever the contract does not reserve to the agencyNon-public agency inputs and outputs may not train publicly or commercially available AI without explicit agency consent

The portability survival scorecard: seven layers, seven ways to break them

Score a data business layer by layer, because portability strips some layers to nothing and barely touches others. For each layer the scorecard asks whether a portability right or exit clause moves it, what a rival would need to rebuild it, and which observable result would prove the layer is not a moat.

A layer survives only if it does not move and passes its disconfirmation test. A business whose surviving layers are all empty, with no aggregate, no labels, no documented authority, no accepted deliveries and no repeat buyers, holds a copy of records its customers can take with them, whatever its record count says.

Read the seven rows from the most portable to the least:

The portability survival scorecard: seven layers, seven ways to break them
LayerMoves under portability?What a rival must rebuildDisconfirmation testIf the test fails
Raw records a customer provided or generatedYes: GDPR Article 20, the Data Act and the CFPB rule all target this layerOne authorization or export per customerA customer's own export reproduces what the buyer needsNo moat; the layer is a copy
Ongoing feed of new recordsOften: Data Act access follows continuing product use, and CFPB authorizations last up to a yearLive authorizations from the same customersA rival with customer authorization gets records as fresh and completeFreshness is shared, not owned
Cross-customer aggregateNo: each person or user moves only their own sliceSeparate consent or contracts from many holdersA few large customers hold most of the records, so a rival needs only themThe aggregate is fragile; value it by its largest customers
Derived labels, outcomes and linksNo: WP242 and the Data Act both leave derived data outThe labeling and linking work, plus the history it ran onRaw records alone match derived records on the buyer's acceptance testThe derived layer adds cost, not defensibility
Documented authority to license for AI useNo: a ported copy serves the person's chosen use, not a licenseFresh, prospective permission from each holderCustomer contracts or notices do not cover licensing to AI developersThe holder has records but nothing it may sell
Record of accepted deliveriesNoTime, and buyers willing to test a newcomerFew deliveries passed a buyer's test without reworkTrust is claimed, not shown
Repeat buyersNo, but often concentratedBuyer relationships and renewalsOne or two buyers account for most contract valueRenewal risk outweighs the moat
Seven moat layers split by a portability line: raw records and the ongoing feed move with the customer, while the cross-customer aggregate, derived labels, licensing authority, accepted deliveries and repeat buyers stay with the holder.
The portability survival scorecard in one view. Editorial framework; a surviving layer still has to pass its test.

Why aggregation and derived labels hold most of the surviving value

They hold it because portability rights are individual and derived data is excluded, so no single exercise of a right can rebuild them. A model developer training a scheduling or claims agent rarely needs one customer's records; it needs patterns across thousands of jobs or cases, each linked to the outcome it produced.

Illustrative: a field-service software company serves 400 business customers and keeps seven years of work orders. One customer uses its contract exit right and exports its own work orders to a rival platform. The rival now holds one slice of 400, without the first-time-fix flag the original platform computes by linking each job to any return visit, because that flag is derived from the raw records rather than provided by the customer. To rebuild the aggregate, the rival needs hundreds of separate exits and the linking work on top.

The same example shows where the layer breaks. If those 400 customer contracts say each customer owns its data and the platform may use it only to provide the service, the aggregate exists but cannot be licensed, and the moat collapses into the authority question. If the 20 largest customers hold most of the work orders, a rival that wins those 20 has most of the coverage. Aggregation is a moat only when it is both licensable and widely spread.

Licensing authority: the layer a ported copy cannot carry

A ported copy gives the recipient data for a purpose the person chose; it does not give a license to train AI on a population's records. That difference is why documented licensing authority usually outlasts portability, and why it is the first layer to verify.

The recipient conditions make the point. Under the CFPB rule, a third party may use authorized financial data only as reasonably necessary for the product the consumer requested, unless the consumer separately authorizes a standalone product. Under the Data Act, data a user shares cannot be used to build a competing connected product. A rival that collects ported records may therefore hold data it is not free to repurpose, while a holder whose contracts already permit licensing keeps an option the rival lacks.

Authority also cannot be manufactured after the fact. In February 2024, FTC staff warned that adopting more permissive data practices, such as using data for AI training, and disclosing the change only through a surreptitious, retroactive amendment to terms or a privacy policy may be unfair or deceptive. An investor should date the clause that grants licensing authority and check that it applied when the records were collected. Buyers screen for the same thing: OpenAI's data partnerships page says it does not want datasets carrying sensitive or personal details, or third-party information.

Compare the surviving layers on measures with stated denominators, so the result can be checked rather than asserted. The first four follow the research method behind this scorecard; the fifth sizes how much of the business sits under a portability right at all.

Use these definitions, and ask for each one by cohort and date:

  • Rebuild cost: the number of separate authorizations, exits or contracts a rival would need to reach the same coverage, plus the share of records held by the 10 largest customers. A high count with low concentration supports the aggregate layer.
  • Legal scope: for each record source, the document that grants licensing authority, its date, and whether it covers AI training, evaluation or both. Count sources with a dated, prospective grant over all sources offered.
  • Technical utility: the score on the buyer's acceptance test for raw records alone against raw records plus derived labels, on the same sample with the same grader. No lift means the derived layer is cost, not moat.
  • Repeat buyer behavior: buyers with a second purchase over buyers with a first, in the same period, plus the share of contract value from the two largest buyers.
  • Portability exposure: the share of records held under a regime with an access or portability right, such as connected-product data in the EU or consumer financial data in the United States.

Is useful data still cheap to substitute or collect directly?

Sometimes, and portability makes the direct route cheaper. If a buyer can ask record holders to send their data through a standard authorization flow, it may not need an intermediary for the raw layer at all, and the moat claim shrinks to aggregation, labels and authority.

Test the substitution case before the moat case. Price the direct route by counting how many holders a buyer would need to recruit and authorize to reach the coverage its task requires. Then check the recipient conditions: under the CFPB rule, a use beyond the product the consumer asked for needs its own authorization as a standalone product, and under the Data Act a competing connected product is off limits. Judge public and synthetic substitutes on the buyer's acceptance test, not on volume.

If a buyer can reach the needed coverage directly, at acceptable cost and on clean terms, the data company's advantage is convenience: valuable, but priced like a service rather than a moat. If direct collection fails on coverage, labels or terms, the surviving layers are doing real work.

Four ways portability could still erode the surviving layers

The scorecard is a snapshot of rules that are still changing, and four developments could move a surviving layer into the portable column or cap its value. Track each one against its dated source.

  • Scope could widen. Today WP242 and the Data Act leave derived data out; a revision or court reading that brings some derived data into scope would weaken the label layer.
  • Rules can reverse in either direction. The CFPB opened a reconsideration of its 2024 rule in August 2025, so a thesis that depends on one version of a rule needs a plan for the other.
  • Buyers can stay concentrated. Reddit's Form 10-K for fiscal 2025 says that "substantially all of the contract value associated with our licensing revenue is derived from two of our partners." A layer that portability cannot touch still carries renewal risk when two buyers set the price.
  • Large buyers push for portability themselves. OMB memorandum M-25-22 tells federal agencies to seek data and model portability, clear licensing and pricing transparency in AI contracts to limit vendor lock-in, which caps switching costs on anything already delivered.

A portability diligence script for any data company, SourceX included

Ask each company which of the seven layers holds its value, and for the evidence at that layer using your definitions. The same script applies to SourceX.

SourceX is one application of the question. It helps companies convert the business records they choose into licensed AI datasets, and its work spans the whole deal: sourcing a company and qualifying it, checking its rights, drawing up the contract, preparing the records, delivering them and settling payment. Each company signs only after approving the scope and terms, and no data is delivered before that signature, so the authority layer is settled before any record moves. Those are process facts, not proof that any layer is defensible.

SourceX has not published supplier, buyer, renewal or concentration metrics; it shares company and commercial metrics with investors directly. Its raise is open only to accredited investors under Rule 506(c); anyone weighing it can ask for investor materials through the investors page. Bring these questions:

  • Which of the seven layers carries most of your value, and which would survive if every customer exported tomorrow?
  • What share of your records sits under a regime with an access or portability right, and how did you measure it?
  • For each record source, which dated document grants licensing authority for AI training, and did it apply when the records were collected?
  • On the same sample and grader, how much does your derived layer add over raw records on a buyer's acceptance test?
  • How many customers hold half of your records, and how many buyers account for half of contract value?
  • Which pending rule change would move one of your layers into the portable column?

How the survival scorecard was built, and what it cannot see

The seven-layer scorecard, the regime table and the five measures are an editorial framework built for this article on 2026-10-26. Evidence comes from the text of GDPR Article 20, the Article 29 Working Party portability guidance, the European Commission's 2022 Data Act impact assessment and its Data Act explainer, the CFPB's 2024 rule and its 2025 reconsideration notice, OMB memorandum M-25-22, FTC staff guidance from February 2024, OpenAI's data partnerships page and Reddit's annual report for fiscal 2025. The field-service company is illustrative and describes no real business.

Not measured: how often customers actually exercise portability rights, the share of any company's value held in each layer, rebuild costs in time or money, and the acceptance-test lift from derived data. Each needs company records that nobody has published. The legal summaries are general information, not legal advice, and the rules described may change.

Frequently asked questions

Does the right to data portability destroy data moats?

It destroys moats built only on holding a copy of records customers provided or generated, because those are what GDPR Article 20, the EU Data Act and the CFPB's 2024 rule let users move. It leaves derived data, cross-customer aggregates, licensing authority, accepted delivery history and repeat buyers in place. Whether a business survives depends on which of those layers carries its value.

Is derived or inferred data covered by GDPR data portability?

Generally no. The Article 29 Working Party's portability guidance excludes data derived or inferred from what a person provided, such as a profile built from raw smart-meter readings, while counting data generated by the person's activity as provided. Counsel should confirm how that line applies to a specific dataset.

What can users take with them under the EU Data Act?

Since 12 September 2025, users of connected products can access the raw and pre-processed data their use generates, free of charge, and have it shared with a third party they choose. Inferred and derived data are out of scope, and data obtained this way cannot be used to develop a competing connected product.

How can an investor test whether a data moat survives portability?

Score each layer of the business on whether a portability right or exit clause moves it, then run a test that could break it: whether a customer's own export reproduces what buyers need, whether raw records match derived ones on an acceptance test, whether contracts permit licensing, and whether buyers renew. Layers that do not move and pass their tests are the moat.

Does SourceX publish portability or renewal metrics?

No. SourceX has not published supplier, buyer, renewal or concentration metrics and shares company and commercial metrics with investors directly. Accredited investors can ask for materials at sourcex.si/investors; an offering, if one is made, is open only to verified accredited investors.

Sources

  • GDPR Article 20 lets a data subject receive personal data they provided in a structured, commonly used and machine-readable format and transmit it to another controller. Source
  • The Article 29 Working Party's portability guidance counts activity-generated data as provided and excludes derived or inferred data. Source
  • The Commission's 2022 Data Act impact assessment judged use of the GDPR portability right largely theoretical and limited to consent- or contract-based processing. Source
  • The EU Data Act, applicable since 12 September 2025, gives connected-product users free access to generated data, excludes derived data and bars use to build a competing product. Source
  • The CFPB rule limits authorized third parties to uses reasonably necessary for the requested product, excludes targeted advertising, cross-selling and data sales, and caps authorization at one year. Source
  • The CFPB opened a reconsideration of its 2024 personal financial data rights rule in August 2025. Source
  • OMB M-25-22 tells agencies to seek data and model portability and clear licensing in AI contracts and bars training commercial AI on non-public agency data without consent. Source
  • FTC staff warned in February 2024 that retroactive, surreptitious terms changes allowing uses such as AI training may be unfair or deceptive. Source
  • OpenAI's data partnerships page says it does not seek datasets with sensitive or personal information or information belonging to a third party. Source
  • Reddit's fiscal 2025 Form 10-K says substantially all of the contract value associated with its licensing revenue is derived from two partners. 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