This chapter identifies the distinctive value proposition of data spaces. It characterises multi-party data sharing as a collective-action problem that neither market pricing nor centralised co-ordination resolves, and defines the boundary conditions under which data spaces hold a comparative advantage over bilateral contracts, centralised platforms and privacy-enhancing technologies. It examines the three mechanisms through which data spaces address this problem: distributed governance, open interoperability and provider-defined usage control. The chapter concludes with the bootstrapping problem that explains much of the gap between investment and operational viability, drawing on case studies of successful and failed initiatives.
Data Spaces between Co‑ordination Challenges, Financial Viability and Public Policy
Abstract
Data spaces target a specific class of data-sharing problem, namely multi-party co‑ordination in sectors where no single participant holds all the data needed for system-level outcomes. Section 3.1 discusses this co‑ordination problem behind multi-party data sharing. Section 3.2 locates data spaces relative to alternative mechanisms – such as bilateral contracts, centralised platforms and services based on privacy-enhancing technologies (PETs) – and defines the boundary conditions under which data spaces have a comparative advantage. Section 3.3 examines the three operational mechanisms through which data spaces address the co‑ordination problem: (i) distributed governance; (ii) open interoperability; and (iii) provider-defined usage control. Section 3.4 turns to a second-order challenge that explains the maturity and adoption gap to a significant extent.
Multi-party data sharing is, at its root, a collective-action problem (Olson, 1965[1]). Multiple parties must co‑ordinate to produce a shared good despite individual incentives that work against collective provision. The two mechanisms that ordinarily co‑ordinate economic activity – market pricing and centralised authority (Williamson, 1985[2]; Williamson, 1991[3]) – both face structural obstacles in this context.
The failure of price formation: in standard markets, prices co‑ordinate allocation without requiring institutional intervention. However, data shared within multi-party ecosystems cannot be efficiently priced because their values are inherently context-dependent.1 This makes standardised valuation across participants structurally difficult (see Box 1.1 of Chapter 1).2 A component traceability record in an automotive supply chain, for instance, has different value to the tier-1 supplier, the original equipment manufacturer (OEM), the recycler and the regulator. The total social value exceeds the sum of bilateral values because all parties having access enables system-level outcomes (supply chain resilience, regulatory compliance, circular economy integration) that no single data transaction captures. Frischmann (2012[4]) argues that resources exhibiting these properties generate social value that price mechanisms alone may not fully capture. This does not imply that market pricing for data is generally unworkable: commercial data markets and data-brokerage services demonstrate that price-mediated exchange functions where data products are sufficiently standardised. The claim here is narrower: in multi-party ecosystems where value is collective, context-dependent and not assignable to a single rights-holder, the conditions for efficient price formation are weakened.
The failure of centralised co‑ordination: where prices cannot help co‑ordinate, a vertically integrated or dominant actor typically steps in, setting the terms and internalising the co‑ordination costs. In multi-party data sharing, this route tends to be blocked wherever the participants are competitors or operating in different jurisdictions and refuse to cede governance authority to a rival-adjacent or otherwise untrusted data intermediary. The most instructive historical case is Covisint, an automotive data exchange platform. Launched in 2000 by General Motors, Ford and DaimlerChrysler (and later joined by Renault and Nissan), Covisint was designed to centralise procurement and supply chain data across the industry. Suppliers resisted, seeing the platform as a tool for buyer leverage over pricing and inventory rather than as neutral infrastructure (Arbin and Essler, 2005[5]; Gerst and Bunduchi, 2005[6]). The auction business was sold off, the founding OEMs withdrew and the original exchange model was abandoned by 2004. The failure was not technological but rather institutional. Various waves of supply chain data-sharing approaches – from electronic data interchange to integration based on service-oriented architecture and, more recently, centralised control towers – each encountered similar governance limitations at a new level of abstraction.3
The compounding of both failures by asymmetric cost-value distribution among participants: in multi-party data sharing, those who bear the costs of contributing data frequently do not capture proportional value from their use (Gelhaar et al., 2021[7]; Brousseau, Toledano and Eustache, 2024[8]). Tier-2 automotive suppliers, for instance, invest in data collection and formatting for traceability records whose primary value accrues to OEMs and regulators; farmers bear the digitisation costs of generating field-level data while downstream processors and technology providers capture most of the analytical value. Practitioners at the OECD Expert Workshop confirmed this imbalance constitutes a persistent challenge (Annex A). The asymmetry is not merely distributional; it is a structural participation barrier that directly affects the second level of the bootstrapping problem analysed in Section 3.4. This is because data providers whose expected returns fall below their contribution costs will rationally under-provide data regardless of the governance model.
For government ministries and public administrations, the absence of price-mediated co‑ordination is the default condition rather than a failure mode. Public-sector data sharing does not operate on a profit-and-loss logic. Rather, the return on investment of a public sector data space is measured in public value generation, such as reduced administrative costs, cross-agency policy coherence and improved delivery of services to citizens (OECD, 2019[9]). This is the limiting case of Frischmann’s (2012[4]) argument above. Where infrastructural resources in commercial settings generate social value that price mechanisms capture only partially, the value generated by public-sector data sharing can almost entirely consist of non-market goods and therefore escapes price formation altogether. The co‑ordination problem can nonetheless persist in this setting. It typically takes the form of institutional fragmentation across ministries with separate mandates, budgets and accountability structures, rather than as failed price formation.4 Section 4.5 and Box 4.2 of Chapter 4 discuss public sector data-sharing frameworks and their public value logic in more detail.
Data space architectures decompose the capabilities needed for multi-party data sharing5 into modular functional components, commonly referred to as “building blocks”. These include components that correspond to the defining features of the narrow definition (Section 2.3 of Chapter 2), such as data exchange infrastructure and usage control. They also include additional capabilities such as publication, discovery and marketplace functions that support the commercial layer through which participants derive value from shared data. The DSSC Blueprint (DSSC, 2026[10]) provides the most widely adopted taxonomy, grouping building blocks into three categories: data interoperability, data sovereignty and trust, and data value creation. The International Data Spaces Association (IDSA) Radar tracks self-reported adoption of these building blocks across initiatives, providing an empirical proxy for which capabilities data spaces prioritise at different maturity stages (Figure 3.1).
The empirical evidence is consistent with these types of failures. The marketplace building block, which embodies price-mediated data exchange, declines from 18% adoption across all IDSA Radar initiatives to 8% at operational stage. This suggests that operational data spaces tend to move away from marketplace logic as they mature (Figure 3.1). As the sample of operational data spaces is small (25 entries), and building block adoption is self-reported (Box 1.1 of Chapter 1), the pattern should be read as indicative rather than conclusive. It is, however, consistent with case study analysis. The data spaces that have achieved higher maturity, such as the Finnish Maritime Data Space (MDS), operate on cost sharing and reciprocal exchange models under neutral governance rather than on data monetisation through a centralised intermediary. They treat shared data as infrastructure to be jointly used and maintained rather than as a commodity to be priced or controlled by a single actor.
Note: Adoption rates are computed by checking whether each building block name appears in the concatenated multi-value string reported by each entry. The denominator is the full population (235 entries for “all” and 25 for “operational”). Building blocks follow the categories defined in the DSSC Blueprint: data interoperability (data exchange, data models and formats, provenance and traceability), data sovereignty and trust (identity management, access and usage policy and control, trust), and data value creation (publication and discovery, data services and offerings descriptions, marketplaces and usage accounting).
Source: Based on IDSA (n.d.), IDSA Data Spaces Radar, www.dataspaces-radar.org.
When market pricing cannot co‑ordinate (data) allocation and no single actor is trusted, the co‑ordination problem identified in Section 3.1 requires an organisational form that is neither purely market-based nor centralised. Williamson (1985[2]; 1991[3]) argues precisely that for transactions characterised by high asset specificity, uncertainty and frequency (all three of which characterise multi-party data exchange via data spaces), “hybrid” governance forms emerge as a distinct organisational response. This “is not a loose amalgam of market and hierarchy but possesses its own disciplined rationale” (Williamson, 1991[3]).6 Data spaces provide one such hybrid governance form (as discussed in Section 3.3.1).
Four specific boundary conditions define the class of problem (use cases) for which data spaces have a comparative advantage over the alternative mechanisms compared in Table 3.1 (e.g. bilateral contracts, centralised platforms and approaches based on PETs. These conditions are (i) three or more parties; (ii) competitive or jurisdictional relationships among them that preclude trusting a single commercial intermediary; (iii) non-excludable collective value that bilateral contracts cannot capture; and (iv) ongoing data flows requiring continuous governance rather than one-off transactions. Where these conditions are absent, simpler data-sharing mechanisms may be sufficient and may not justify the co‑ordination costs that data spaces entail.
Note: IAM = Identity and Access Management Platform; DPP = Digital Product Passport; EU GDPR = General Data Protection Regulation; IDS = Industrial Data Space; ODS = Open Data Spaces; ToS = Terms of Service. The table compares idealised properties of each mechanism along dimensions relevant to the boundary conditions identified in Section 3.2. The data spaces column reflects general properties of the narrow definition; specific implementations vary across architectural approaches (e.g. IDS-RAM/EDC, ODS-RAM). Where illustrative examples are given, they are not exhaustive.
The alternative mechanisms in Table 3.1 each serve legitimate purposes where one or more of these boundary conditions is absent, although none addresses all four simultaneously.
Bilateral contracts are the default mechanism for pairwise data sharing, but their co‑ordination costs scale combinatorially with the number of participants [n(n-1)/2 for n parties], making them impractical where boundary condition (i) applies. They also provide no shared governance layer through which collective value, boundary condition (iii), can be created or allocated.
Centralised commercial platforms have built successful multi-party data exchange businesses. Snowflake, AWS Data Exchange and Databricks together serve tens of thousands of customers; roughly a quarter of Snowflake’s customers maintain sustained cross-account data-sharing relationships. Many organisations evidently accept vendor terms of service in exchange for functionality, speed and network effects. These platforms succeed by offering a different trade-off, forgoing data provider control and multi-stakeholder governance in exchange for lower co‑ordination costs and immediate functionality.
PETs address the confidentiality dimension of multi-party data sharing, enabling computation on combined data without revealing individual inputs. They do not, however, address the institutional co‑ordination problem of who governs, who pays and how collective value is allocated (see Section 3.3.3 for how PETs complement rather than substitute for data space governance).
Grouping 147 IDSA Radar initiatives by their self-reported value propositions yields a pattern consistent with the boundary conditions identified above. The five domains where data spaces have a partially unique structural advantage (supply chain visibility across competitors, health data exchange across jurisdictions, multimodal mobility co‑ordination across operators, energy system integration across prosumers and utilities, and smart city data sharing across public and private providers) all share the structural feature identified above: they are n-party co‑ordination problems where no commercial participant is trusted to serve as intermediary. Figure 3.2 provides a quantitative complement, showing the operational share by sector. The sectors with the highest operational share tend to align with the domain clusters identified above.
Note: “Operational conversion rate” is the number of IDSA Radar entries that have reached operational status as a proportion of all entries in that sector. Sectors are self-reported and multi-valued: a single entry may appear in more than one sector, so sector totals exceed 235. “Agriculture” combines entries labelled “Agriculture/Agrifood” (14) and “Agriculture” (7) in the Radar. Only sectors with three or more total entries are shown.
Source: Based on IDSA (n.d.), IDSA Data Spaces Radar, www.dataspaces-radar.org.
Data spaces address the multi-party co‑ordination problem through three operational mechanisms. These operational mechanisms are in line with the narrow definition (Section 2.3 of Chapter 2) and their ordering below reflects how adoption proceeds in practice, informed by case study evidence:
Distributed governance (Section 3.3.1) provides the institutional foundation.
Open interoperability (Section 3.3.2) provides the technical infrastructure that participants actually deploy.
Provider-defined usage control embedded as a capability in the interoperability stack. The trust capability (Section 3.3.3), comprising in particular identity and data usage control, distinguishes data space interoperability from both commercial platform interoperability and traditional industry consortium data exchange.
The distinctive feature is not any single mechanism but their integration across three facets commonly distinguished in the dataspace literature (i.e. technical, business and legal). Through this integration, governance rules become technically enforceable through the interoperability stack; contractual usage conditions become machine-readable and portable across the ecosystem; and business arrangements are anchored to a shared institutional framework. No alternative data-sharing mechanism integrates institutional, technical and contractual controls into a single operational framework as tightly as data spaces.
Participants at the Expert Workshop highlighted the potential that data spaces offer to address challenges in cross-border use cases, amid increased risks of barriers to cross-border data flows (Annex A). In particular, they highlighted that data space architecture allowed participants to negotiate machine-readable policies (at the level of data usage rights). These can be configured to be compatible with the regulations and laws of multiple jurisdictions simultaneously, making these conditions available to all ecosystem participants through the standardised protocol. Where bilateral contracts would require renegotiating terms for each regulatory combination and each two parties, data spaces can in principle encode cross-jurisdictionally compatible requirements once and propagate them across the ecosystem. In so doing, data spaces significantly reduce transaction costs.
IDSA Radar data underscore the difficulty of activating all three mechanisms simultaneously. Of the 153 initiatives at implementation and operational stages, only 45 score on all three operational mechanisms (distributed governance, open interoperability and usage control) (see Figure 3.1). The rarity of multi-mechanism deployment is consistent with the argument that each mechanism addresses a distinct dimension of the co‑ordination problem, and partial activation limits the value a data space can deliver.7
3.3.1. Distributed governance
Despite the significant importance of distributed governance, in every mature data space examined for this paper, a neutral intermediary or orchestrator occupies a key governance role: a state enterprise (Fintraffic), a non-profit association (Catena-X Automotive Network e.V., see Box 3.1) or an industry consortium (BoostAeroSpace). The pattern is aligned with Guggenberger, Schlueter Langdon and Otto (2025[11]), who characterise data spaces as “meta-organisations” in which legally independent entities voluntarily co‑ordinate under a shared framework. Concentrating governance in a single commercial intermediary would convert the arrangement into a hierarchy, potentially destroying the voluntary co‑ordination that defines it. Even data ecosystems that start on decentralised principles tend to establish central institutions as they scale (Kernstock et al., 2025[12]), but these institutions must remain structurally neutral.
The drift towards centralisation reflects well-documented co‑ordination failures: as the network grows, the costs of distributed decision making (rule-setting, dispute resolution, onboarding) rise faster than its benefits. Consequently, individual participants face incentives to free-ride on shared infrastructure rather than contribute to it. A neutral central institution lowers these transaction costs and concentrates accountability, but at the price of reintroducing the very governance asymmetry that decentralised principles were designed to avoid. This structure enables competitors to share data where no single participant is trusted as data intermediary, or accommodates public-interest data flows.
The neutral intermediary or orchestrator also need to understand both the overall value proposition and the addressable market of the data space in order to initiate and sustain collective action. In particular it needs to address how each stakeholder tends to only understand its own business model in relation to the data space while missing a view of the overarching business value. However, distributing the benefits created by a data space requires complex negotiation among members, including orchestrators who may also have their own business interests (Annex A). The governance challenge is therefore not simply to ensure neutrality in the abstract, but rather to design institutional arrangements that harness stakeholders’ incentives without allowing them to capture disproportionate value. The neutrality requirement of the EU Data Governance Act (DGA) can be understood, in part, as a regulatory response to this tension, although it is under revision (see Section 4.1 of Chapter 4).
Where price formation for data fails (Section 3.1), distributed governance can provide an alternative resource allocation mechanism. In this context, rather than pricing individual data transactions, participants collectively negotiate how to share the costs of maintaining shared infrastructure and how to distribute the value it generates. The Finnish MDS illustrates this most directly: its value proposition is cost sharing (quantifiable reductions in fuel costs, port fees and emissions penalties) rather than data monetisation. Lessons learnt from BoostAeroSpace show a similar logic through membership-based cost allocation. Both treat shared data as infrastructure to be collectively maintained, consistent with Frischmann (2012[4]).
Catena-X Automotive Network originated from industry initiatives in 2020-2021. It was established as the association Catena-X Automotive Network e.V. on 15 May 2021 by 17 German founding members, including ARENA2036 and the Fraunhofer Institutes and Research Units (Fraunhofer-Gesellschaft) alongside 15 companies (Catena-X, 2023[13]). As of November 2024, there were 193 members. Catena-X as a data ecosystem has three parts. The first is the Catena-X Automotive Network e.V., which is responsible for standards, certifications and governance. Members contribute to developing the data ecosystem, including its standards, and are obliged to comply with the latter. The second part is the development environment, which proposes candidate standards and develops open-source reference implementations. Finally, the operating environment offers various open-source and commercial services and applications.
The values offered by Catena-X are reflected in its “10 Golden Rules” (Catena-X, 2025[14]), which are highlighted as contributing to the bigger pictures of European data space strategies: (i) values of Gaia-X and IDSA; (ii) data sovereignty; (iii) governance framework; (iv) freedom of contract; (v) registration and certification; (vi) compliance by conduct; (vii) compliance by design; (viii) security; (ix) control; and (x) legal provisions and disclaimer. In particular, Catena-X is presented as an effective means to ensure compliance with data- and sector-specific regulations, improve supply chain management, and enable secure and efficient data sharing. These values are expected to be further enhanced as more participants join Catena-X through network effects.
Industry-owned consortia such as SWIFT8, SITA9 and GS110 have co‑ordinated multi-party data exchange across competing firms for decades without the data space architecture. What data spaces add is a technical enforcement layer (including provider-defined, machine-readable usage control that travels with data) and provenance infrastructure that makes such control verifiable (Section 3.3.3). BoostAeroSpace illustrates the trajectory: established in 2011 as a governance consortium for aerospace supply chain data exchange, it subsequently adopted data space components (Eclipse Dataspace Components, IDSA architecture) to add technical enforcement to its institutional framework.
An illustrative case is EONA-X, an operational data space in the mobility sector that brings direct competitors under a neutral data intermediary for sustainability data. The Finnish MDS, led by Fintraffic (a state-owned enterprise with an indefinite time horizon), demonstrates business-model-first design: technology choices follow from business requirements.
The IDSA Radar data reflect a related pattern: EU funding accounts for about 46% of implementation-stage entries but drops to about 20% at operational stage (Box 1.1 of Chapter 1). Two interpretations are consistent with this pattern: data spaces that reach operational maturity may develop funding models independent of public grants, or EU-funded initiatives may be more likely to fail before reaching operational stage. The cross-sectional data cannot distinguish between these dynamics, and both may operate simultaneously. Either way, the pattern suggests that EU grant funding alone is not sufficient to sustain data spaces through to operational viability.
The case studies above illustrate what these models look like in practice: membership-based cost allocation (BoostAeroSpace), cost sharing tied to measurable operational savings (Finnish Maritime, see Box 3.2 and Annex A),11 or industry association funding (EONA-X). Notably, none of these models relies on data commercialisation. The decline in marketplace building block adoption from 18% across all initiatives to 8% at operational stage (Figure 3.1) suggests that data monetisation through pricing and trading has not proven a viable funding model for operational data spaces. The implications of this funding transition for policy design are discussed in Section 4.2 of Chapter 4.
3.3.2. Open interoperability and portability
If participants must adopt a proprietary technology stack to exchange data, the infrastructure provider risks becoming a de facto central authority. Standards-based protocols significantly reduce this risk of capture and ensure participants retain the ability to exit, switch and connect across domains.
Practitioner accounts at the OECD Expert Workshop suggest that interoperability is the primary reason participants adopt data space components. Organisations deploy data space components to solve a concrete problem: exchanging data across organisational boundaries without rebuilding integrations for each new partner. The governance framework (Section 3.3.1) determines why participants join; the interoperability stack determines what they actually deploy. Usage control capabilities (Section 3.3.3) come embedded in this stack, available when needed but not the primary adoption motivation in many cases.
Data spaces specify data exchange through standardised protocols that define a common transaction process (with a possible negotiation phase as foreseen in the Eclipse Dataspace Protocol or optionally provided in the Open Data Spaces Protocol). Reference implementations provide interoperable interfaces (e.g. connectors) and standardised information models ensuring semantic interoperability (see Box 2.1 of Chapter 2). Emerging AI-based semantic mediation may, over time, reduce reliance on rigid standardised information models by translating across heterogeneous schemas at runtime. However, the governance and provenance guarantees required in regulated dataspaces still favour explicitly agreed models. Open interoperability keeps switching costs low and makes investment in data products portable across vendors. It also enables multi-domain value propositions that proprietary platforms cannot support without bilateral agreements with every other platform.
The Data Space for Traffic Measurements, an operational initiative in the Flanders region, illustrates standardised protocols by unifying more than 500 entities through a linked data standard; interoperability is the primary value mechanism. Catena-X deploys Eclipse Dataspace Components and standardised information models across the entire automotive value chain, making usage control portable across the supply chain. A policy attached at the tier-2 supplier level travels with data through to the OEM because the underlying protocol is standardised.
Adoption data show consolidation around open standards within data spaces. At the operational stage, about a third use EDC, while the FIWARE connector drops from about 20% overall to none at operational stage. This pattern is consistent with consolidation around the protocol stack that supports the full Eclipse Dataspace Protocol negotiation and transfer state machine. However, it may also reflect a reporting bias: IDSA maintains the IDSA Radar, and initiatives using alternative connector technologies may be less likely to register (see Box 1.1 of Chapter 1).
The pattern also illustrates the role of open-source implementations as a de facto standardisation mechanism in this space. The Eclipse Dataspace Components and FIWARE codebases are themselves the reference points around which interoperability converges. This lowers the cost of multi-vendor adoption relative to purely specification-led standards.
Architectural decoupling is another route to open interoperability with the ODS-RAM a notable example. The architecture of ODS-RAM is organised around three pillars: ontology and semantic interoperability; data addressability and discoverability; and identity and usage control. Each combines technical and rights-based functions as loosely coupled, independently replaceable components (IPA, 2026[15]; Tsuda, 2026[16]).12 This pairing allows the ontology layer to evolve independently of the data layer. In this way, semantic models can be updated without re-engineering data pipelines or breaking interfaces. In addition, the pairing supports incremental, opt-in adoption – “detachability” in Tsuda (2026[16]). This is directly relevant to the bootstrapping problem discussed in Section 3.4. Emerging approaches extend this principle through “dynamic ontologies”, which use large language models to automate mappings across heterogeneous ontologies.13
The practical consequence is that switching costs can be reduced not only across vendors implementing the same protocol, but also across architectural components. An organisation can replace its identity mechanism or semantic model without re-engineering its data transfer pipeline. The IPA-Catena-X proof of concept demonstrated product carbon footprint data exchange between ODS-RAM- and EDC-based implementations for EU Battery Regulation compliance. In so doing, it provides early evidence that protocol- and architecture-level interoperability is achievable (IPA and Catena-X, 2025[17]). (See Section 4.3 of Chapter 4).
Commercial platforms show signs of moving in a similar direction at the protocol layer. Databricks open-sourced its Delta Sharing protocol and Unity Catalog, suggesting the economic logic of open standards can override the lock-in incentive even for commercial actors. This is not convergence on data space architecture: Delta Sharing does not implement the Eclipse Dataspace Protocol or the Open Data Spaces Protocol. Moreover, open-sourcing a protocol does not entail adopting distributed governance or provider-defined usage control. It does, however, narrow the gap on the interoperability dimension. This means the structural differentiation of data spaces increasingly rests on governance neutrality (Section 3.3.1) and the trust capability (Section 3.3.3). It is the integration of these dimensions with the open interoperability layer that constitutes the comparative advantage. Governance neutrality is available through public bodies, usage control through bilateral contracts, and open standards through industry consortia. However, no alternative mechanism combines all three.
3.3.3. Identity, usage control and provenance for trust
The trust capability embedded in the architecture is the key distinction between data space interoperability on the one hand, and commercial platform interoperability and traditional industry consortium data exchange on the other. Provider-defined data usage control travels with data across the ecosystem, supported by identity verification and provenance tracking. These mechanisms presuppose underlying digital security guarantees (OECD, 2022[18]): (i) confidentiality (shared data are not disclosed beyond authorised recipients); (ii) integrity (data and related rules are not tampered with in transit or at rest); and (iii) availability (the data and the infrastructure are reliably accessible). Without these guarantees, access and usage control mechanisms lack credibility regardless of how they are specified and implemented. Implementations range from machine-readable policies (enforced, for example, by connectors) to other mechanisms such as PETs (e.g. federated learning, multi-party computation and differential privacy).
Identity and access control are necessary but not distinctive. All data-sharing mechanisms provide verifiable identity and authorisation: centralised platforms authenticate users through proprietary identity and access management systems; bilateral contracts rely on legal identification clauses; services based on PETs authenticate participants before granting access to secure enclaves. Data spaces implement the same function through verifiable credentials (IDS-RAM) or OpenID Connect (ODS-RAM). As the key difference, identity is federated or decentralised rather than managed by a single operator. This matters for governance neutrality, since no single party controls the identity layer. However, it does not by itself distinguish data spaces from other mechanisms.
A highly distinctive capability is data usage control: provider-defined, machine-readable or otherwise technically supported conditions on what may happen to data after access is granted e.g. purpose limitation, time bounds, geographic restrictions, prohibition of onward sharing. On centralised platforms, the operator’s terms govern post-access use; in bilateral contracts, usage conditions lack technical enforcement; in secure computation, control is limited to computation-only use cases. A comparative of data spaces is the ability to provide user-defined, technically enforceable access and usage control across a multi-party ecosystem.
This capability addresses a barrier documented across sectors, private and public alike. Regulatory requirements, including data protection regulations, play a critical role in safeguarding individuals and preventing the misuse of data. However, possibly inappropriate interpretations of these requirements can overly constrain data accessibility even where sharing would be lawful and beneficial. Organisations are thus often reluctant to share without assurance that robust safeguards operate in the receiving entity (OECD, 2025[19]). By making conditions such as purpose limitation explicit, machine-readable and auditable, usage control gives data providers, and the officers responsible for their compliance, demonstrable grounds for approving sharing that over-cautious interpretation would otherwise block.
The primary mechanism proposed for data usage control, in particular by the IDSA-RAM, is through machine-readable conditions expressed in the Open Digital Rights Language (ODRL),14 negotiated through the Eclipse Dataspace Protocol and enforced by connectors.15 This faces an inherent limitation: once data leave the provider’s infrastructure, enforcement depends on the data user’s compliance. In practice, data-sharing agreements complement technical controls. To that end, they provide a contractual fallback enforceable through legal channels when technical enforcement fails or proves infeasible. PETs address this limitation by shifting from controlling access to controlling what can be learnt (Box 3.2).
Architectural variants illustrate different implementation strategies and, in the case of usage control, different design philosophies. The IDS-RAM/Eclipse Dataspace Protocol model mandates the full stack, including a mandatory negotiation layer even for fully permissive contracts (i.e. “no operation” policies that impose no restrictions). The ODS-RAM makes the contracting module optional and does not restrict the technical means for usage control. It provides only interfaces that accommodate diverse methods – from contract negotiation to PETs such as trusted execution environments and differential privacy (Tsuda, 2026[16]). The rationale is that binding usage control to a single, institutionally derived protocol can create a bottleneck for market adoption. Thus, while the IDS-RAM approach prescribes the mechanism and achieves adaptability through functional configuration, the ODS-RAM prescribes only the interface and delegates the choice of mechanism to participants.
Provenance and audit are key for making data usage control credible despite the post-transfer enforcement gap. The data user’s access interface (e.g. connector or other protocol-level endpoints) is both the enforcement mechanism and the potential point of circumvention. Consequently, usage control policies alone cannot prevent a determined data user from extracting and misusing data outside the enforcement path. Provenance tracking and audit logs serve as the compensating mechanism. While they do not prevent agreement violations, they make violations more detectable, shifting the problem from technical prevention to deterrence through accountability.
Privacy-enhancing technologies (PETs) enable the collection, processing, analysis and sharing of information, while safeguarding data confidentiality and privacy (OECD, 2023[20]; OECD, 2025[19]). The literature identifies four PET categories as particularly relevant to data spaces:
Federated learning enables model training across participants without data leaving the provider’s infrastructure. This has been demonstrated in healthcare settings where institutions across multiple countries collaboratively trained diagnostic models while preserving privacy.
Secure multi-party computation allows participants to jointly compute functions over combined data without any party revealing individual inputs.
Trusted execution environments, often considered as the foundation of confidential computing, provide hardware-isolated enclaves where data can be processed in decrypted form but cannot be observed or extracted by external software. Closely related architectures, such as Data Control Rooms, package these enclaves with policy enforcement and audit capabilities into a managed environment in which approved analyses only run on the provider's infrastructure.
Synthetic data generation produces artificial datasets that preserve the statistical properties of original data while containing no real records. This enables participants to share representative data for model development, testing or benchmarking without exposing sensitive inputs.
Recent literature suggests that combining PETs with the contract-based framework is more effective than deploying either in isolation. In this approach, a data space connector negotiates a contract that specifies not only who may access data and for what purpose, but through which PET the data must be processed. This makes the choice of privacy technology an enforceable element of the data-sharing agreement. The provider no longer needs to trust the recipient’s compliance because the technology structurally limits what can be done with the data. PET adoption in data spaces does, however, face the same collective-action challenges identified throughout this section: additional computational costs, specialised expertise requirements and the need for co‑ordination among participants on which technologies to deploy (Auñón et al., 2024[21]).
The adoption data reflect this shift from prevention to deterrence. While 35% of all initiatives have adopted the access and usage policy building block, only 28% have done so at operational stage. Meanwhile, the provenance building block is moving in the opposite direction – from 25% to 48% (see Figure 3.1). The divergence becomes intelligible when usage control is understood as a capability embedded in the interoperability stack rather than a stand-alone adoption driver. Participants deploy connectors for interoperability and usage control comes bundled. Provenance must be deliberately adopted as data spaces mature and the stakes of policy violation increase.16 The capability to enforce usage conditions, even when not actively exercised, functions as embedded option value. This is a theoretically grounded hypothesis, not an empirically established finding.
Data spaces may address the first-order co‑ordination problem identified in Section 3.1, but establishing them faces a second-order collective-action challenge of its own, commonly referred to as the bootstrapping problem.17 Data space components, including connectors, semantic models and trust frameworks, are costly to build, but – once designed – they can be re-used by each new entrant.18 The shared infrastructure layer exhibits club good properties: high fixed costs, low marginal costs for each additional deployment and benefits that are non-rivalrous within the ecosystem but excludable to non-participants (Brousseau, Toledano and Eustache, 2024[8]). Individual participants, however, may face onboarding costs (e.g. adapting internal systems, implementing connectors, mapping data to shared semantic models) that can be substantial, particularly for small and medium-sized enterprises (SMEs) (DSSC, 2025[22]).
Initiatives such as the Dataspace Adoption Forum (DSAF) have emerged to help organisations navigate onboarding costs.19 The bootstrapping challenge therefore operates at two levels: the collective investment in shared infrastructure must precede the network effects that justify it; and each participant’s private onboarding investment must be justified by expected returns that may not materialise until the ecosystem reaches sufficient scale.20
The free-riding problem identified by Olson (1965[1]) applies directly, since participants who can benefit from the shared infrastructure without contributing to its establishment have no rational incentive to invest in its provision. Brousseau, Eustache and Toledano (2024[8]) argue that multi-sided data ecosystems without a dominant key actor face the structurally hardest co‑ordination problem, requiring neutral intermediation that commercial platforms cannot provide. Bisière et al. (2025[23]) formalise the resulting chicken-and-egg dynamic: data spaces need participants to create value, but participants will not join without existing value. Moreover, the neutrality requirements of the EU DGA further constrain the cross-subsidisation strategies that commercial platforms use to bootstrap participation (Rochet and Tirole, 2003[24]). The bootstrapping problem is accordingly invoked to justify public investment in data space establishment (Bisière et al., 2025[23]; Otto and Tardieu, 2025[25]) (see Chapter 4).
Frischmann (2012[4]), however, argues that free riding does not necessarily reduce the investment incentives of contributing participants. A participant who benefits from shared infrastructure without contributing proportionally is not undermining the investment case if the contributing participants capture enough private value from their own participation to justify their costs regardless.21 This implies that bootstrapping is a binding constraint specifically when the expected private returns to anchor participants fall below their investment threshold despite positive collective returns (i.e. when no subset of participants can justify the investment from their own expected returns alone).
The more recent economics literature identifies several strategies for overcoming the chicken-and-egg dynamic (Caillaud and Jullien, 2003[26]; Evans and Schmalensee, 2010[27]). Such strategies include attracting participants who can act on both sides of the market (as both data providers and consumers). This would help reduce the need to build each side independently. Instead existing platforms or ecosystems can be leveraged to converge activities into a common infrastructure; high-value initial use cases can demonstrate tangible benefits to early adopters (a “seeding strategy”); “marquee users” can be secured whose participation signals credibility and attracts others; and one side of the market can subsidise the other during the early adoption phase.
The OECD Expert Workshop confirmed that data space practitioners draw on this repertoire (Annex A). A further approach involves keeping the data space operator deliberately compact by connecting “collaborative services” (shared infrastructure guided by public guidelines) with “competitive services” offered by third-party providers through open application programming interfaces (APIs). This approach limits the operator’s own cost structure while enabling a broader ecosystem of value-added services (Annex A). The bootstrapping threshold is further inflated by a recurring co‑ordination cost that practitioners describe as “reinventing the wheel” (Annex A).
Data spaces are frequently developed by consortia whose members vary from one initiative to another, bringing different champions, expertise and views. Excessive time spent debating technical components from scratch is a recurring pattern, slowing development and increasing costs. Baron, Meniere and Pohlmann (2014[28]) analyse a structurally similar phenomenon in technology standardisation, where formal processes entail duplicative research and development, as well as delays due to vested interests. A key benefit of industry consortia is precisely to reduce these costs by settling conflicts of interest upfront. The data space ecosystem, with multiple overlapping initiatives (IDSA, Gaia-X, European Data Spaces Support Centre, sector-specific consortia) each developing partially overlapping building blocks, faces analogous risks.
The Finnish MDS overcame the bootstrapping problem (Box 3.3). Fintraffic, a state-owned enterprise with an institutional mandate, serves as anchor participant. Competitors (shipping lines, port operators) achieved quantifiable cost reductions in fuel, port fees and emissions penalties thanks to their participation in the data space. Operational trials of the Virtual Port Arrivals use case between Lulea and Oxelosund achieved an average 24% reduction in CO2 emissions and 20% reduction in energy consumption (ELS Shipping, n.d.[29]). Extrapolated to all Finnish maritime traffic, the projected annual savings amount to EUR 45 million in fuel and costs related to the EU Emissions Trading System (2022 baseline), as well as 135 000 metric tonnes of avoided CO2 emissions (Siili Solutions, n.d.[30]).22 A concrete revenue-sharing model distributes these savings among the shipping company, the charterer, the port and the system operator (Annex A). The private return to participants exceeds their participation cost; bootstrapping was not a binding constraint because the anchor institution and the quantifiable value proposition together cleared the investment and participation threshold.
Fintraffic is a state-owned enterprise in Finland tasked with the management and development of the nation’s traffic control services. Through its Maritime Data Space (MDS) project, Fintraffic develops practical solutions for sharing data generated by ships. The initiative aims to optimise traffic, reduce fuel consumption, lower emissions and address numerous logistical challenges within ports. At the same time, it seeks to improve safety by sharing real-time information about ship movements, weather conditions and other critical safety factors.
Contrary to previous maritime logistics initiatives that focussed on specific company issues, MDS aims to deliver value to the entire ecosystem. Cost sharing among various industry players is a key value proposition centred around two use cases to optimise ship scheduling and fuel efficiency, thereby reducing delays, costs and environmental impacts. In Just-In-Time Arrivals, data analytics optimise a ship’s journey and port visit by considering factors such as weather, cargo specifics and timings for arrivals, departures and turnarounds. Overall, it aims to enhance the cost effectiveness of shipping operations and the use of port resources. For its part, Virtual Port Arrivals aims to optimise the arrival process of ships at ports by allowing ships to adjust their arrival speed in real time based on the port’s current use rates.
A key factor contributing to the financial viability and success of MDS has been attributed to its focus on governance, business models and value creation prior to technological implementation. The OECD Expert Workshop identified five sequential steps in this approach (Annex A): (i) identify ecosystem champions from shipping companies, charterers, ports, cargo owners and relevant public entities; (ii) develop systemic service design by mapping stakeholders’ challenges and incentives, their roles and operations, data holdings and flows, and existing contractual frameworks; (iii) map the real-life contractual model and value model; (iv) build the business model, including a revenue-sharing arrangement among participants; and (v) enable the solution through data space technology. This process took approximately eight months. The emphasis on demonstrating tangible monetary savings alongside environmental benefits proved essential for securing industry buy-in: financial advantages and environmental compliance aligned in the same direction.
Agdatahub, which declared bankruptcy in November 2024 (Agdata, 2024[31]), represents the opposite case of the bootstrapping problem (Box 3.4). The value of shared agricultural data remained speculative; no anchor participant had a private return sufficient to justify the investment and participation costs on a continuous basis. The data space seemed to have been designed around funding availability rather than an identified business problem. When public grant funding from the French government expired, participation could not sustain itself.
The OECD Expert Workshop highlighted a structural variable: the agricultural sector’s fragmentation into many SMEs made it inherently harder to organise into a data-sharing ecosystem than sectors with concentrated global companies sharing common supply chains (Annex A). This is consistent with the economics literature suggesting that large, fragmented groups face stronger free-rider problems than small, concentrated ones; the individual payoff from collective provision decreases with group size, while the cost of organising voluntary contributions increases.23
In markets dominated by SMEs, regulatory requirements may not serve as an effective incentive for participation. SMEs often need to digitise their operations first, and additional regulatory compliance requirements only increase costs and reduce their capacity to innovate with shared data (Annex A). Similar digitalisation constraints apply beyond firms, such as to government ministries and public administrations (including municipalities) whose operations are often not yet fully digitised.
Founded in 2017 by the French agricultural sector with the support of technical institutes and chambers of agriculture, Agdatahub launched Agritrust around 2020. It received significant additional funding from the French government’s France 2030 initiative (Agdata, 2024[31]). Agdatahub operated as a trusted data intermediary and data space operator for the French and European agricultural sectors, aiming to facilitate secure data sharing between farms and providers of agricultural technologies and services. Its initial goal was “to provide the 380 000 farms in France and all the relevant actors in the sector with innovative tools that take into account the consent of farmers to the use of their data” (Avril, 2023[32]).
Agdatahub’s business model focussed on providing digital tools for managing agricultural data while ensuring compliance with European regulations. This included two flagship offerings (Avril, 2023[32]): Agritrust and API-Agro. Agritrust was conceived as a solution for managing farmers’ consent to the use of their data, based on “the very first agricultural digital identity”. Focusing on digital service providers for farms and available as a paid subscription, Agritrust was also provided as a free mobile application for farmers. API-Agro was a platform for multi-stream exchange between holders and users of agricultural data. The subscription revenue model (including a first level free of charge) allowed relevant sectoral actors to access numerous functionalities and exchange data securely.
Both solutions were designed to meet the needs of the agricultural industry in terms of data dissemination for innovations; to address data control and transparency requirements weighing on farmers; and to respect EU regulations, including the EU DGA and Data Act. Despite its key objective to promote innovation for private stakeholders on the agriculture market, Agdatahub stated its business model proved unsustainable without continued government funding.
BoostAeroSpace occupies an intermediate position as a governance consortium that illustrates lessons relevant to data spaces. Major original equipment manufacturers in the aerospace sector captured enough private return from supply chain efficiency gains to anchor the investment, even as smaller suppliers free-rode on the shared infrastructure. The trajectory from founding (approximately 2011) to financial viability (approximately 2019) holds two key lessons: bootstrapping can be overcome through sufficient anchor participant returns; and the time horizon required can be considerably longer than a typical public grant cycle of two to four years. This has implications as discussed in Chapter 4.
A common pattern emerges across the three cases. The actual value in data spaces is created at the level of individual use cases, not at the level of the infrastructure itself (Annex A). The infrastructure layer provides the common governance, interoperability and trust framework within which use cases operate. However, participants invest because of the concrete value they expect to derive from specific applications. In the cases discussed, and confirmed more broadly at the OECD Expert Workshop across practitioners from Europe, Japan and North America, prioritising use-case value over infrastructure development was consistently associated with stronger prospects for financial sustainability (Annex A). The Finnish MDS designed its business model around two specific use cases before selecting technology. ABtC followed a similar pattern in Japan, anchoring the ecosystem around a concrete initial use case (carbon footprint data for battery compliance) before expanding scope. Agdatahub invested in a general-purpose infrastructure platform without demonstrated use-case value sufficient to sustain participation. The distinction between use-case business models and infrastructure business models is therefore analytically consequential for the policy analysis in Chapter 4.
The three cases differ systematically along two structural dimensions that prior work on data-sharing ecosystems identifies as critical for viability: (i) whether participants hold symmetric or asymmetric positions in the value chain; and (ii) whether a “keystone actor”24 co‑ordinates the ecosystem (Iansiti and Levien, 2004[33]; Eustache, Brousseau and Toledano, 2025[34]).
Table 3.2 situates the three cases discussed above, together with the Ouranos battery passport use case (Chapter 4), within a 2×2 matrix. The configurations align with the bootstrapping outcomes observed: where a keystone actor absorbs initial integration costs and asymmetries are managed through redistribution (BoostAeroSpace; Ouranos battery passport), bootstrapping is more readily overcome. Where no keystone is present and participants are heterogeneous in capabilities and incentives (Agdatahub), the orchestrator must perform both strategic alignment and technical intermediation, and the bootstrapping problem is most acute.
Source: Author's elaboration based on Eustache, Brousseau and Toledano (2025[34]), The Economics of Data-Sharing: An Empirical Investigation of Data Sharing Ecosystems, https://hal.science/hal-05542222v1; and case studies from Sections 3.3 and 3.4.
A distinct revenue model is emerging from IT service providers, including telecommunications operators and potentially cloud platforms. In this model, data space connectivity is offered as an incremental managed service on top of existing infrastructure rather than building a stand-alone intermediary that must reach its own critical mass (Box 3.5). The model sidesteps the bootstrapping problem by leveraging infrastructure and customer relationships that already exist at scale. It illustrates that commercial viability in data spaces may be achieved by operators with existing infrastructure, neutral intermediary positioning and volume-based revenue that scales with adoption. Whether integrated platform operators can credibly serve this role depends on two factors: the regulatory neutrality requirements discussed in Section 4.1 of Chapter 4; and whether participants will trust an intermediary that also competes for their compute, analytics or AI workloads with their competitively sensitive data.
T-Systems, the IT services subsidiary of Deutsche Telekom, treats data space connectivity as a managed service comparable to Internet or voice connectivity. Its Data Intelligence Hub offers commercialised products for ecosystem onboarding, sandbox testing, identity verification (based on Gaia-X Digital Clearing House standards) and geo-information analytics. T-Systems was the first organisation to receive IDSA connector certification (January 2024) and serves as a certified onboarding service provider for Catena-X. Revenue is product-based rather than tied to data transactions. The model builds on existing network infrastructure, making data space connectivity an incremental service rather than a stand-alone investment requiring its own critical mass.
The model has a cross-border dimension. In September 2024, T-Systems and the NTT Group (NTT Communications, renamed NTT DOCOMO Business in July 2025, with NTT DATA) deployed a data space test environment in Japan. This integrated Gaia-X trust anchor mutual recognition between Europe and Japan, building on a 2022 NTT–Catena-X partnership announcement (NTT, 2022[35]; 2025[36]). IDSA established a Global Telecom Community in June 2024 with T-Systems, NTT Communications (now NTT DOCOMO Business) and KPN (Netherlands) as founding members. NTT DATA has set a target of approximately EUR 310 million in data space-related revenue by 2030, one of the few publicly stated commercial benchmarks in the ecosystem.
References
[31] Agdata (2024), “Agdatahub demande l’ouverture d’une procédure collective – Agdatahub”, https://agdatahub.eu/en/agdatahub-procedure-collective-liquidation/ (accessed on 25 February 2025).
[45] Andresel, M. et al. (2024), “Adapting ontology-based data access for data spaces”, The Second International Workshop on Semantics in Dataspaces, co-located with the Extended Semantic Web Conference, May 26 – 27, 2024, Hersonissos, Greece: SDS 2024 co-located with ESWC 2024, https://ceur-ws.org/Vol-3705/paper06.pdf.
[5] Arbin, K. and U. Essler (2005), “Covisint in Europe: Analysing the B2B auto e-marketplace”, International Journal of Automotive Technology and Management, Vol. 5/1, p. 31, https://doi.org/10.1504/ijatm.2005.006878.
[44] Arrow, K. (1962), “Economic Welfare and the Allocation of Resources for Invention”, in The Rate and Direction of Inventive Activity, National Bureau of Economic Research, Princeton University, https://doi.org/10.1515/9781400879762-024.
[21] Auñón, J. et al. (2024), “Evaluation and utilisation of privacy enhancing technologies – A data spaces perspective”, Data in Brief, Vol. 55, p. 110560, https://doi.org/10.1016/J.DIB.2024.110560.
[32] Avril (2023), “Agdatahub raises €4.8 million to boost the development of a French and European agricultural infratech in a trusted framework”, https://presse.avril.com/agdatahub-raises-e4-8-million-to-boost-the-development-of-a-french-and-european-agricultural-infratech-in-a-trusted-framework/?lang=en (accessed on 25 February 2025).
[28] Baron, J., Y. Ménière and T. Pohlmann (2014), “Standards, consortia, and innovation”, International Journal of Industrial Organization, Vol. 36, pp. 23-35, https://doi.org/10.1016/j.ijindorg.2014.05.004.
[43] Battaglini, M. and T. Palfrey (2024), “Organizing for collective action: Olson revisited”, Journal of Political Economy, Vol. 132/9, https://doi.org/10.1086/729580.
[23] Bisière, C. et al. (2025), “The economics of data spaces”, Policy Paper, SE Digital Center and Toulouse School of Economics, France, https://www.tse-fr.eu/sites/default/files/TSE/documents/DigitalCenter/policy_paper/the_economics_of_data_spaces_july_2025_policy-paper.pdf.
[8] Brousseau, E., J. Toledano and L. Eustache (2024), “Economics of data sharing: Study on the emergence and creation of value within data-sharing ecosystems”, Position Paper, Paris Dauphine University / Gaia-X Institute, https://gaia-x.eu/wp-content/uploads/2024/03/Study-on-the-emergence-and-creation-of-value-within-data.pdf.
[26] Caillaud, B. and B. Jullien (2003), “Chicken & egg: Competition among intermediation service providers”, The RAND Journal of Economics, Vol. 34/2, pp. 309-328, https://doi.org/10.2307/1593720.
[14] Catena-X (2025), “Catena-X: 10 golden rules”, https://catenax-ev.github.io/docs/next/regulatory-framework/30000ft/ten-golden-rules (accessed on 10 April 2026).
[13] Catena-X (2023), “About Us”, https://catena-x.net/en/about-us (accessed on 10 April 2026).
[10] DSSC (2026), “Data Spaces Blueprint v3.0”, https://dssc.eu/space/BVE/357074549/Data+Spaces+Blueprint+v3.0 (accessed on 9 April 2026).
[22] DSSC (2025), “Participant adoption and sustainability in data spaces”, https://dssc.eu/publications (accessed on 16 April 2026).
[29] ELS Shipping (n.d.), Virtual Arrival, https://www.eslshipping.com/en/services/virtual-arrival (accessed on 14 October 2026).
[34] Eustache, L., E. Brousseau and J. Toledano (2025), “The economics of data-sharing: An empirical investigation of data sharing ecosystems”, Forty-Sixth International Conference on Information Systems (ICIS), Nashville, Tennessee, https://hal.science/hal-05542222v1.
[27] Evans, D. and R. Schmalensee (2010), “Failure to launch: Critical mass in platform businesses”, Review of Network Economics, Vol. 9/4, https://doi.org/10.2202/1446-9022.1256.
[4] Frischmann, B. (2012), Infrastructure: The Social Value of Shared Resources, Oxford University Press.
[7] Gelhaar, J. et al. (2021), “Towards a taxonomy of incentive mechanisms for data sharing in data ecosystems”, PACIS 2021 Proceedings, ACIS 2021 Proceedings 121, https://aisel.aisnet.org/pacis2021/121/.
[6] Gerst, M. and R. Bunduchi (2005), “Shaping IT standardization in the automotive industry – the role of power in driving portal standardization”, Electronic Markets, Vol. 15/4, pp. 335-343, https://doi.org/10.1080/10196780500302872.
[11] Guggenberger, T., C. Schlueter Langdon and B. Otto (2025), “Data spaces as meta-organisations”, European Journal of Information Systems, Vol. 5/34, pp. 822-842, https://doi.org/10.1080/0960085X.2025.2451250.
[42] Hart, P. and C. Saunders (1997), “Power and trust: Critical factors in the adoption and use of electronic data interchange”, Organization Science, Vol. 8/1, https://doi.org/10.1287/orsc.8.1.23.
[33] Iansiti, M. and R. Levien (2004), “The Keystone advantage: What the new dynamics of business ecosystems mean for strategy, innovation, and sustainability”, Academy of Management Perspectives, Vol. 20/2, https://doi.org/10.5465/AMP.2006.20591015.
[41] IDSA (2024), IDS Reference Architecture Model, Version 4.0, https://github.com/International-Data-Spaces-Association/IDS-RAM_4_0.
[15] IPA (2026), “Open Data Spaces Reference Architecture Model (ODS-RAM)”, https://open-dataspaces.gitbook.io/ods-docs/ods-ram/v2 (accessed on 1 April 2026).
[17] IPA and Catena-X (2025), “Catena-X and Ouranos ecosystem successfully demonstrate data space interoperability”, https://catena-x.net/news/catena-x-and-ouranos-ecosystem-successfully-demonstrate-data-space-interoperability/ (accessed on 6 March 2026).
[12] Kernstock, P. et al. (2025), “Establishing and governing data ecosystems at the crossroads of centralization and decentralization”, Electronic Markets, Vol. 35/71, https://doi.org/10.1007/s12525-025-00810-x.
[40] Löhe, J. and C. Legner (2010), “SOA adoption in business networks: do service-oriented architectures really advance inter-organizational integration?”, Electronic Markets, Vol. 20/3-4, pp. 181-196, https://doi.org/10.1007/s12525-010-0046-7.
[36] NTT (2025), “Renewal of NTT Group’s corporate identity”, https://group.ntt/en/newsrelease/2025/05/09/250509z.html (accessed on 4 May 2026).
[35] NTT (2022), “NTT Com and NTT DATA to develop data-sharing eco-system that interconnects with Europe’s Catena-X Platform to protect data sovereignty will enable secure sharing of automotive, aviation & energy data in global value chains”, https://group.ntt/en/newsrelease/2022/05/26/220526b.html (accessed on 4 May 2026).
[19] OECD (2025), “Sharing trustworthy AI models with privacy-enhancing technologies”, OECD Artificial Intelligence Papers, No. 38, OECD Publishing, Paris, https://doi.org/10.1787/a266160b-en.
[20] OECD (2023), “Emerging privacy-enhancing technologies: Current regulatory and policy approaches”, OECD Digital Economy Papers, No. 351, OECD Publishing, Paris, https://doi.org/10.1787/bf121be4-en.
[46] OECD (2022), “Measuring the value of data and data flows”, OECD Digital Economy Papers, No. 345, OECD Publishing, Paris, https://doi.org/10.1787/923230a6-en.
[18] OECD (2022), OECD Policy Framework on Digital Security: Cybersecurity for Prosperity, OECD Publishing, Paris, https://doi.org/10.1787/a69df866-en.
[9] OECD (2019), Public Value in Public Service Transformation: Working with Change, OECD Publishing, Paris, https://doi.org/10.1787/47c17892-en.
[1] Olson, M. (1965), The Logic of Collective Action: Public Goods and the Theory of Groups, Harvard University Press.
[25] Otto, B. and H. Tardieu (2025), Data Sharing: Closing the Gap in the Draghi Report’s Call for AI Competitiveness in Europe, Le Grand Continent.
[24] Rochet, J. and J. Tirole (2003), “Platform competition in two-sided markets”, Journal of the European Economic Association, Vol. 1/4, pp. 990-1029, https://doi.org/10.1162/154247603322493212.
[30] Siili Solutions (n.d.), “Virtual port arrival - maritime innovation”, Siili Solutions, https://www.siili.com/cases/virtual-port-arrival-maritime-innovation (accessed on 10 April 2026).
[39] Staebler, M., F. Koester and C. Schlueter-Langdon (2023), “Towards solving ontological dissonance using network graphs”, Americas Conference on Information Systems, p. 2023, https://arxiv.org/abs/2308.14326v1.
[38] Swink, M. et al. (2024), “Supply chain visibility types and contextual characteristics: A literature‐based synthesis”, Journal of Business Logistics, Vol. 45/1, pp. E-12366, https://doi.org/10.1111/jbl.12366.
[16] Tsuda, M. (2026), Why Open Dataspaces: Design Philosophy and Architectural Paradigm, Information-technology Promotion Agency, Tokyo, https://www.ipa.go.jp/en/digital/architecture-guidelines/individual-link/p1o1lf000001xv4n-att/WhyOpenDataspaces_en.pdf.
[3] Williamson, O. (1991), “Comparative economic organization: The analysis of discrete structural alternatives”, Administrative Science Quarterly, Vol. 36, pp. 269-296, https://doi.org/10.2307/2393356.
[2] Williamson, O. (1985), The Economic Institutions of Capitalism: Firms, Markets, Relational Contracting, The Free Press, New York.
[37] World Wide Web Consortium (2018), “ODRL Information Model 2.2”, W3C, https://www.w3.org/TR/odrl-model/ (accessed on 23 April 2026).
← 1. The problem is more fundamental than context-dependency alone. Arrow (1962[44]) identified the information paradox: a prospective buyer cannot assess the value of information without first possessing it, yet once disclosed the information has effectively been transferred. Price negotiation requires revealing the very asset being sold, undermining the conditions for market exchange. In multi-party settings, this paradox is compounded because the same data have different value to each participant, making even bilateral price discovery intractable at ecosystem scale.
← 2. The value depends on what other data it is combined with (combinatorial value), on uses that cannot be fully specified ex ante (option value) and on downstream applications whose beneficiaries extend beyond the immediate transaction (demand-side externalities) (OECD, 2022[46]).
← 3. Electronic data interchange (EDI) in the 1980s, supplier portals in the 1990s, application programming interfaces and integration platforms in the 2000s, and centralised control towers in the 2010s all enabled data exchange where one dominant party set the terms. This was not the case in multilateral settings where governance had to be shared among participants with competing interests. EDI adoption was characterised by buyer coercion and power asymmetry: dominant firms used EDI to extend control over supply chain partners, and small firms adopted primarily because of external pressure from powerful trading partners rather than perceived mutual benefit (Hart and Saunders, 1997[42]). Integration based on Service-Oriented Architecture in the 2000s reproduced hub-and-spoke governance dominated by focal organisations despite the technology’s promise of dynamic, decentralised business networks; polycentric arrangements between equal partners remained rare (Löhe and Legner, 2010[40]). Supply chain control towers in the 2010s achieved visibility only within the focal firm’s immediate trading partners; governance structure, not technology, is the binding constraint on multi-tier visibility (Swink et al., 2024[38]).
← 4. Two features of this setting sharpen the problem. Statutory mandates bind each authority’s data to defined purposes, so that inter-governmental re-use can require a new legal basis rather than a commercial agreement. In addition, the payoff structure of public administration can institutionalise risk aversion, since officials may face sanctions for wrongful sharing while capturing little of the value that sharing creates.
← 5. The multi-party framing also covers internal data flows within a single global enterprise: cross-country, cross-subsidiary or cross-business-line transfers typically involve semi-autonomous departments, branches or subsidiaries that face the same trust, control and interoperability problems as independent organisations, and increasingly turn to data space architectures to address them.
← 6. The asset-specificity leg requires qualification. Data themselves are nonrival and technically redeployable across relationships at near-zero marginal cost. The specificity lies not in the data but in the participation investment: adapting internal systems; implementing connectors; mapping data to shared semantic models; and building governance relationships are sunk costs whose value is largely specific to the ecosystem (Section 3.4 of Chapter 3). Context-dependent data value reinforces this specificity, since data that are technically portable may yield substantially less value outside the ecosystem where complementary datasets and institutional infrastructure give it meaning. They are, however, appropriate only for a subset of data-sharing problems.
← 7. Catena-X illustrates the interdependency (Box 3.1 of Chapter 3). Distributed governance through Catena-X Automotive Network e.V. establishes the institutional foundation; open interoperability via EDC and IDSA standards provides the technical infrastructure through which competitors actually exchange data; and usage control (the trust capability embedded in the interoperability stack) makes competitors willing to contribute proprietary supply chain data. Regulatory requirements (provenance documentation and identity management for EU Battery Regulation compliance) amplify the urgency (see Chapter 4).
← 8. Society for Worldwide Interbank Financial Telecommunication (SWIFT) operates the global messaging network for financial transactions.
← 9. Société Internationale de Télécommunications Aéronautiques (SITA) provides IT and communications infrastructure for the air transport industry.
← 10. GS1 manages global supply chain standards including barcodes and electronic data interchange.
← 11. In the Finnish Maritime Data Space, quantifiable operational savings (reduced fuel costs, port fees and emissions penalties) are distributed among participants (shipping company, charterer, port and system operator) according to a collectively negotiated formula. In this way, the data space is funded through the value it generates rather than through external grants (see Box 3.3 of Chapter 3 and Annex A).
← 12. Each pillar operates independently and is loosely coupled, so participants can selectively implement components based on their use case and maturity rather than adopting the full stack at once. The architecture emphasises “incremental payoff for incremental investment” and “Minimal Yet Viable” design, drawing on the Classical Dataspaces principle that solutions should “not exist only as monolithic solutions”. (Tsuda, 2026[16]). See also IPA (2026[15]).
← 13. Ontology matching based on large language models has become an active research area, with dedicated benchmarks (OAEI-LLM, OAEI-LLM-T) and systems such as MILA reporting F1 scores of 0.83‑0.95 on biomedical alignment tasks. Within the data spaces context, early work addresses the “ontological dissonance” across participating domains (Staebler, Koester and Schlueter-Langdon, 2023[39]) and adapts ontology-based data access for data spaces (Andresel et al., 2024[45]). Within ODS-RAM v2, a “dynamic ontology” concept has been proposed as a design direction (Tsuda, 2026[16]). Deployment in operational data spaces remains experimental.
← 14. ODRL is a W3C Recommendation (World Wide Web Consortium, 2018[37]) that provides a standardised vocabulary for expressing permissions, prohibitions and obligations over digital content and services. In the data space context, ODRL policies specify conditions such as permitted purposes, temporal validity, geographic restrictions, and onward-sharing prohibitions in a machine-readable format that connectors can interpret and enforce automatically, without requiring human review of each data-sharing request.
← 15. The Dataspace Protocol defines a state machine for contract negotiation and data transfer between a provider and a data user connector. The Eclipse Dataspace Components, the most widely deployed reference implementation, operationalises this protocol: the provider connector publishes a catalogue of available data assets with attached ODRL policies; the data user connector initiates a negotiation; once both sides agree on terms, the protocol transitions to a transfer state in which data flow only under the negotiated conditions. Enforcement occurs at the connector level: the provider’s connector verifies that the data user has accepted the negotiated conditions before authorising data transfer. Post-transfer enforcement, however, depends on the data user’s connector honouring the agreed policies. The IDS-RAM acknowledges that this requires trusted platforms and cannot prevent circumvention by administrators with infrastructure access (Section 4.1.6 of IDS-RAM 4.0) (IDSA, 2024[41]).
← 16. The combination of ODRL policies, connector enforcement and provenance tracking therefore functions as a socio-technical enforcement system: technical controls at the point of transfer, complemented by accountability mechanisms that make post-transfer non-compliance detectable and legally actionable.
← 17. The bootstrapping problem in multi-sided ecosystems is analysed in the platform economics literature as a chicken-and-egg dynamic (Caillaud and Jullien, 2003[26]).
← 18. The Eclipse Dataspace Components reference implementation is developed once as open-source software, although each participant must deploy and configure their own instance (managed-service offerings increasingly reduce this burden for smaller firms). An industry-specific semantic model (such as Catena-X’s automotive data models) is negotiated once and re-used across the supply chain. A trust framework’s certification process, once designed, applies uniformly to new entrants.
← 19. These costs reflect the technical complexity of data space infrastructure itself, which some practitioners have described as a barrier to adoption in its own right. Where a need for data sharing involves few parties, bilateral APIs can offer a cheaper and more pragmatic alternative.
← 20. The concept of “dataspaces” introduced by Franklin et al. eliminates this problem by design, treating integration as a private investment decision in which each participant pays only for connections whose value exceeds their cost. Modern data spaces invert that logic, requiring collective infrastructure to precede individual participation.
← 21. As Frischmann argues, “there is a mistaken tendency to believe that any gain or loss in profits corresponds to an equal or proportional gain or loss in investment incentives, but this belief greatly oversimplifies the decision making process and underlying economics” (Frischmann, 2012[4]).
← 22. These figures are based on business calculations from real ESL Shipping voyage data and port call data. The EUR 45 million figure is a counterfactual projection (what would have been saved had the system been operational across Finnish maritime traffic in 2022), not measured aggregate savings. The 24% and 20% reduction figures are from ESL Shipping’s operational data on a specific route, not independently verified. No peer-reviewed publication was identified at time of writing.
← 23. Battaglini and Palfrey (2024[43]) formalise this insight using mechanism design. They show that for large groups the probability of successful collective action converges to zero, although optimal organisational mechanisms can slow the rate of decline. No empirical study was identified that applies Olson’s group-size logic specifically to data space participation across sectors; the workshop observation represents practitioner experience rather than a tested empirical finding.
← 24. See Iansiti and Levien (2004[33]) who argue that business ecosystems contain different kinds of actors, which they classify into: (i) Keystones: actors that occupy hubs in the network, set standards and actively shape ecosystem health by increasing value for many others (e.g. Microsoft for the PC ecosystem in the 2000s, Walmart for retail supply chain); (ii) Dominators: actors that exploit their hub position to extract value rather than share it, and often kill the ecosystem they sit in; and (iii) Niche players: specialised actors that depend on the keystone but contribute distinct value.