Data sharing across organisations and jurisdictions generates economic and social value, increasingly so as progress in artificial intelligence depends on high-quality, domain-specific data. This chapter sets out why data spaces have emerged as a response to the trust and collective-action problems that centralised platforms leave unresolved, and why few of the more than 200 registered initiatives have reached operational viability despite significant public investment. It frames the question the report addresses, namely under what conditions data spaces provide value that alternative data-sharing mechanisms cannot, and describes the evidence base and its limitations.
Data Spaces between Co‑ordination Challenges, Financial Viability and Public Policy
Abstract
Data sharing generates economic value through multiple channels: it strengthens supply chain resilience by enabling firms to anticipate disruptions; raises productivity by reducing duplicative data collection; accelerates innovation by allowing novel combinations of datasets that no single organisation could assemble alone; and more generally provides the diverse, domain-specific training data on which artificial intelligence (AI) systems increasingly depend (OECD, 2019[1]; Jones and Tonetti, 2020[2]; OECD, 2025[3]). These benefits are well established. Indeed, the policy case for enhancing data access and sharing has been articulated across successive OECD legal instruments, most recently the OECD Recommendation on Enhancing Access to and Sharing of Data [OECD/LEGAL/0463] (OECD, n.d.[4]).
The urgency of this agenda is sharpening. As the pace of AI capability gains from scaling training data and compute alone appears to be slowing down, attention is shifting towards data quality, curation, and domain specificity as the binding constraints on further progress (Hoffmann et al., 2022[5]; Zeff, 2024[6]; Lu, 2025[7]). Publicly available data from the open web may be abundant, but they are often noisy, biased or legally contested. Organisations already exploit their own operational records, sensor streams, clinical observations and supply chain metrics for internal AI applications, including through retrieval-augmented generation and domain-specific model fine-tuning. What they lack are the governance frameworks, legal certainty and control mechanisms to reduce the risks of data sharing, and often a clear view of the economic returns. This is particularly true where combining complementary datasets from multiple holders could unlock substantially greater value, but that value accrues collectively rather than to any single participant.
The commercial data platform market provides supportive evidence of this economic value. Combined revenues of two of the leading data sharing platforms exceeded USD 10 billion in 2025, serving over 30 000 customers including most of the Fortune 500 firms.1 These platforms operate data marketplaces and cross-organisation exchange protocols that reach thousands of data recipients across competing ecosystems. The commercial data platform market has demonstrated that organisations will share data when the infrastructure is convenient, the network effects are strong and the value proposition is immediate.
Yet certain use cases impose conditions that centralised platforms may be unable to meet. For instance, firms will not share competitively sensitive data (e.g. supply chain positions, production quality metrics, emissions profiles) unless they trust that usage conditions will be enforced and that governance arrangements do not unfairly advantage one participant over others. For many use cases, these limitations may not be significant: the platform’s functionality, speed and network effects might outweigh the governance constraints. For others, however, the limitations may be binding constraints. This may be especially true where the value of shared data is collective and infrastructural in character, generating benefits that extend beyond any single transaction and organisation (Frischmann, 2012[8]). The questions become: which problems fall into each category, and is the distinction sufficiently sharp to guide investment and public policies?
Data spaces have emerged as one of the principal data governance responses to the trust and collective-action problems that centralised platforms leave unresolved. These collective-action problems, which involve multi-party data sharing among possible competitors, across jurisdictions and under data control requirements, are not specific to any one region. They arise wherever digitalisation, supply chain integration or AI development require controlled access to distributed data. The largest concentration of data space initiatives has emerged within the European policy context. This has been driven by the European Strategy for Data (European Commission, 2020[9]), the EU Data Governance Act (European Union, 2022[10]), the EU Data Act (European Union, 2023[11]) and associated public investment programmes (see Chapter 4). However, data spaces are gaining policy recognition and adoption around the world, sometimes via functionally equivalent arrangements under different names (see Chapter 2).
Outside Europe, Japan has advanced the furthest. The Ouranos Ecosystem, an industrial initiative initiated and supported by Japan's Ministry of Economy, Trade and Industry (METI), draws on the Open Data Spaces (ODS) technical architecture developed by Japan’s Information-technology Promotion Agency (IPA). It has produced operational deployments such as the Automotive and Battery Traceability Centre (ABtC) (see Chapters 2 and 3). Other countries such as Canada and Korea, as well as Brazil and the People’s Republic of China (hereafter “China”), are at earlier or accelerating stages. Outside the data spaces label, India’s Data Empowerment and Protection Architecture (DEPA) and the Trusted Exchange Framework and Common Agreement (TEFCA) in the United States for health data are examples of functionally equivalent arrangements.
More recently, the United States launched the Genesis Mission in November 2025 (The White House, 2025[12]). This directs the creation of a federated data-sharing platform across the 17 US national laboratories to accelerate AI-enabled research in priority sectors, including nuclear fusion, quantum computing, advanced manufacturing, energy infrastructure and critical materials. Initial investments by the US Department of Energy include approximately USD 320 million announced in December 2025 to build AI infrastructure (the American Science Cloud and the Transformational AI Models Consortium). This was followed by a USD 293 million Request for Applications in March 2026 inviting interdisciplinary teams to address the programme’s identified national science and technology challenges (US Department of Energy, 2025[13]; US Department of Energy, 2026[14]). While the programme does not adopt the data spaces label, it functionally creates a federated data-sharing model. This, in turn, highlights the importance of intellectual property frameworks for contributed proprietary training data, an issue addressed in Section 3.3 of Chapter 3.
The International Data Spaces Association (IDSA) Radar,2 the largest publicly available registry of data space initiatives, lists 235 entries as of February 2026 across sectors ranging from mobility and health to agriculture and energy. The IDSA Radar nearly tripled in size from 87 entries in 2022 to 235 in early 2026, confirming wide-scale experimentation and growing maturity of data spaces. Yet the ecosystem remains overwhelmingly pre-operational. Of the 235 initiatives, only 25 have reached operational status. None have achieved what the IDSA Radar classifies as “scaling”, the only stage that IDSA’s definitions explicitly associate with financial and operational viability (IDSA, 2024[15]) (Figure 1.1).3 Only one in nine entries has reached operational status, a “valley of death”4 that constitutes the defining challenge facing the ecosystem (Box 1.1). This warrants particular attention given that public investment in data space initiatives in the European Union alone is estimated at over EUR 1.7 billion as of early 2025 (Otto and Tardieu, 2025[16]).5
This gap between investment and operational maturity has raised a pointed question among policymakers around the world: are data spaces worth the investment? This paper reframes that question in more precise terms. The issue is not whether data spaces create value in the generic sense that any data-sharing infrastructure creates value; commercial platforms have already demonstrated that proposition. Rather, the question is whether certain data-sharing problems require the specific properties that data spaces provide, in particular federated architecture, provider-defined usage control, distributed governance and open interoperability. If such a class of problems exists, then the gap between investment and operational scaling identified above is worth closing, and policy attention may need to focus on how to close it most effectively. If it does not, then other approaches including centralised platforms are the rational default, and the policies directed towards data spaces would deserve re-evaluation.
Note: Development stages are self-reported by data space operators to the IDSA Radar and may not be updated as initiatives progress or fail. Entries are predominantly European in origin, reflecting the historical concentration of data space activity around EU-funded programmes (see Box 1.1). “Implementation” denotes a detailed project plan and resources for a pilot but no operational use case. “Operational” denotes at least one use case with data flowing between providers and recipients. One entry classified as “Scaling” on the IDSA Radar dashboard is recorded without a development stage in the underlying dataset export; the figure therefore shows zero scaling entries.
Source: Based on IDSA (n.d.), IDSA Data Spaces Radar, www.dataspaces-radar.org.
This paper provides an evidence-based assessment of the distinctive value proposition of data spaces. In particular, it asks under what conditions, and through what mechanisms, do data spaces solve problems that alternative data-sharing approaches cannot.
This focus on value precedes, and is a necessary condition for, the question of business models. Business model analysis asks how an organisation creates, delivers and captures value (Osterwalder and Pigneur, 2010[17]). As applied to data spaces, authors such as Mertens et al. (2024[18]) decompose this into the use cases that generate value; the value proposition offered to participants; the implementation model; and the balance between cost structure and revenue. However, this decomposition presupposes that the value in question has been identified. If it is misidentified, business models built on that assumption risk being structurally misaligned with the value the data space may produce. A data space whose value is collective and infrastructural, for instance, may not be able to sustain business models designed around transactional data monetisation.
IDSA Radar data are consistent with this pattern: data spaces that adopted marketplace-based revenue models tend to struggle to reach higher degrees of maturity. That said, this should be read as indicative rather than conclusive given the relatively small sample of operational data spaces (25 entries) and their self-reporting nature (Box 1.1). The present paper therefore addresses two questions: what distinctive value do data spaces create, and through which mechanisms? Only once this foundation is established can the question of how to capture that value through viable business models be answered soundly.
To mitigate the data source limitations highlighted in Box 1.1, the empirical base of this draft report combines the quantitative data from the IDSA Radar with case study evidence from, and international comparisons with, parallel developments in Japan (ABtC, the ODS ecosystem), India (DEPA), and the United States (TEFCA for health data). The analysis also draws on findings from the OECD Workshop on Assessing the Sustainability of Business Models in Data Spaces and the Role of Public Policies (hereafter “the Expert Workshop”). Held on 11 March 2025 at the Data Spaces Symposium in Warsaw, the workshop brought together practitioners, policymakers and industry experts from Europe, Japan and North America. Together, they critically examined the financial sustainability of data space business models on the basis of concrete use cases, both successful and failed (Annex A).6
The IDSA Radar is a voluntary, self-reported registry whose registered initiatives are predominantly European, reflecting both a registration bias and the European concentration of data space activity noted in Section 1.1. Where this paper draws on IDSA Radar statistics, the quantitative patterns describe the European data spaces ecosystem as registered in the IDSA Radar, not necessarily the global landscape of federated data-sharing arrangements.
Beyond this coverage bias, the Radar presents other specific data quality constraints. Missing data rates are high: approximately 50% of entries lack geographical focus, 58% lack funding source information and 78% lack start dates. Development stages are recorded at a point in time and may not reflect current status; initiatives that have failed may still appear as operational. Analyses of subsets (funding, governance, access models) are based on the reporting population only.
This measurement gap is itself a key finding. Most operational data spaces do not track the metrics needed to assess their own value. The Radar captures no outcome data: no transaction volumes, no participant growth trajectories, no cost comparisons, no productivity indicators, and no measures of data flows enabled or withheld. Case studies provide qualitative evidence of mechanism activation but were not designed as controlled evaluations. Consequently, they cannot isolate the causal contribution of data space architecture from confounding factors such as regulatory incentives, pre-existing industry relationships or the maturity of participating firms’ digital infrastructure.
The measurement problem has structural roots. Data spaces are, by design, distributed systems in which no single entity observes all transactions. The very data provider control principles that define data spaces make centralised data collection difficult. Participants may be reluctant to share performance data with a registry for the same reasons they are reluctant to share operational data with centralised platforms. The result is a paradox: the data-sharing infrastructure that data spaces provide is itself under-measured because the principles that make it distinctive also make it hard to observe.
A deeper issue compounds this difficulty: Measuring the value of data. The value of data is not intrinsic: it depends on context, governance, and the extent to which it is shared and linked with complementary datasets. No single valuation method captures the full picture (OECD, 2022[19]; Bisière et al., 2025[20]). In particular, data use and sharing generate spillovers, both positive and negative, that are not reflected in market prices. These challenges apply even more severely to data spaces, where value is predominantly collective and infrastructural (Chapter 3): avoided costs, system-level resilience, ecosystem-wide regulatory compliance and innovation spillovers from unanticipated data combinations all resist conventional quantification. The proposed DGP Measurement Agenda [DSTI/DPC/DGP(2026)4] aims to address this broader gap.
The paper is structured as follows. Chapter 2 establishes a two-tier working definition for data spaces that distinguishes broad (architecture-agnostic) from narrow (federated, with provider control) formulations. Chapter 3 identifies the distinctive value proposition of data spaces by analysing the co‑ordination problem they address; the boundary conditions under which, and operational mechanisms through which, they hold a comparative advantage over alternative data-sharing mechanisms; and the bootstrapping challenge that explains the maturity and adoption gap to a significant extent. Chapter 4 provides a brief overview of the emerging policy landscape, examining regulatory drivers, public investment design and cross-border interoperability efforts across jurisdictions. Chapter 5 draws conclusions and sets out policy considerations related to data space initiatives.
References
[20] 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.
[26] Data Spaces Symposium (2025), “Memories of Data Spaces Symposium 2025”, https://www.data-spaces-symposium.eu/program-2025 (accessed on 1 March 2026).
[25] Databricks (2026), “Databricks grows >65% YoY, surpasses $5.4 billion revenue run-rate”, https://www.databricks.com/company/newsroom/press-releases/databricks-grows-65-yoy-surpasses-5-4-billion-revenue-run-rate (accessed on 9 February 2026).
[9] European Commission (2020), Communication from the Commission to the European Parliament, the Council, the European Economic and Social Committee and the regions: A European strategy for data, European Commission, Brussels, https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX%3A52020DC0066.
[11] European Union (2023), Regulation (EU) 2023/2854 on harmonised rules on fair access to and use of data (Data Act), European Union, Brussels.
[10] European Union (2022), Regulation (EU) 2022/868 of the European Parliament and of the Council of 30 May 2022 on European data governance and amending Regulation (EU) 2018/1724 (Data Governance Act), European Union, Brussels, https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX%3A32022R0868.
[24] Frank, C. et al. (1996), “Surviving the “valley of death”: A comparative analysis”, The Journal of Technology Transfer, Vol. 21, pp. 61-69, https://doi.org/10.1007/bf02220308.
[8] Frischmann, B. (2012), Infrastructure: The Social Value of Shared Resources, Oxford University Press.
[5] Hoffmann, J. et al. (2022), “Training compute-optimal large language models”, arXiv, Vol. 35/2203.15556v1, https://arxiv.org/abs/2203.15556v1.
[15] IDSA (2024), IDS Reference Architecture Model, Version 4.0, https://github.com/International-Data-Spaces-Association/IDS-RAM_4_0.
[23] IDSA (n.d.), “International Data Spaces”, https://internationaldataspaces.org/ (accessed on 31 March 2026).
[2] Jones, C. and C. Tonetti (2020), “Nonrivalry and the economics of data”, American Economic Review, Vol. 110/9, pp. 2819-2858, https://doi.org/10.1257/aer.20191330.
[7] Lu, C. (2025), “The race to efficiency: A new perspective on AI scaling laws”, arXiv 2501.02156v3, https://arxiv.org/abs/2501.02156v3.
[22] Markham, S. et al. (2010), “The valley of death as context for role theory in product innovation”, Journal of Product Innovation Management, Vol. 27/3, pp. 402-417, https://doi.org/10.1111/j.1540-5885.2010.00724.x.
[18] Mertens, C. et al. (2024), “Data spaces business models”, Position Paper, Zenodo (CERN European Organization for Nuclear Research), https://doi.org/10.5281/zenodo.14101303.
[3] OECD (2025), Enhancing Access to and Sharing of Data in the Age of Artificial Intelligence, OECD Publishing, Paris, https://doi.org/10.1787/23a70dca-en.
[19] 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.
[1] OECD (2019), Enhancing Access to and Sharing of Data: Reconciling Risks and Benefits for Data Re-use across Societies, OECD Publishing, Paris, https://doi.org/10.1787/276aaca8-en.
[27] OECD (2013), Commercialising Public Research: New Trends and Strategies, OECD Publishing, Paris, https://doi.org/10.1787/9789264193321-en.
[4] OECD (n.d.), Recommendation of the Council on Enhancing Access to and Sharing of Data, OECD/LEGAL/0463, https://legalinstruments.oecd.org/en/instruments/OECD-LEGAL-0463 (accessed on 6 March 2023).
[17] Osterwalder, A. and Y. Pigneur (2010), Business Model Generation: A Handbook for Visionaries, Game Changers, and Challengers, Wiley, New York.
[16] 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.
[21] Snowflake (2026), “Financial results for the fourth quarter and full-year of fiscal 2026”, https://www.snowflake.com/en/news/press-releases/snowflake-reports-financial-results-for-the-fourth-quarter-and-full-year-of-fiscal-2026/ (accessed on 28 April 2026).
[12] The White House (2025), “Launching the Genesis mission”, https://www.whitehouse.gov/presidential-actions/2025/11/launching-the-genesis-mission/ (accessed on 28 April 2026).
[14] US Department of Energy (2026), “Energy department announces $293 million in funding to support genesis mission national science and technology challenges”, https://www.energy.gov/articles/energy-department-announces-293-million-funding-support-genesis-mission-national-science (accessed on 28 April 2026).
[13] US Department of Energy (2025), “Energy department advances investments in AI for science”, https://www.energy.gov/articles/energy-department-advances-investments-ai-science (accessed on 28 April 2026).
[6] Zeff, M. (2024), “Current AI scaling laws are showing diminishing returns, forcing AI labs to change course”, TechCrunch, https://techcrunch.com/2024/11/20/ai-scaling-laws-are-showing-diminishing-returns-forcing-ai-labs-to-change-course/ (accessed on 16 March 2026).
← 1. Snowflake reported USD 4.7 billion in revenue for fiscal year 2026, ending 31 January 2026 (Snowflake, 2026[21]), while Databricks reached an annualised revenue of USD 5.4 billion as of January 2026 (Databricks, 2026[25]).
← 2. The IDSA Radar is a self-reported, voluntary registry maintained by the International Data Spaces Association (IDSA), a not-for-profit association based in Germany with more than 140 member organisations, which “brings together global members from both industry and research” to “develop and promote the concept of data spaces” (IDSA, n.d.[23]).
← 3. Over half of all entries (128, or 54%) are at the implementation stage (project plan and resources but no operational use case). Only 25 (11%) have reached operational status. Public funding dominates the early stages (46.1% of implementation-stage data spaces received EU funding). However, it recedes sharply at maturity (20% of operational ones), and no market-based revenue model has taken its place: only 2 of 96 initiatives reporting an access model charge a fee. The ecosystem is not merely young; it faces a structural viability gap.
← 4. The “valley of death” refers to the critical gap between early-stage development and self-sustaining operation, during which initiatives are most vulnerable to failure. The concept originates in technology commercialisation literature, where it describes the funding and capability gap between research output and market viability (Frank et al., 1996[24]; Markham et al., 2010[22]). The OECD has applied the concept to science-based innovation policy, identifying the valley of death as the phase where public research funding has ended but private investment has not yet materialised (OECD, 2013[27]).
← 5. The EUR 1.7 billion figure (EUR 1 735.4 million) originates from a funding portfolio compiled by the Gaia-X European Association for Data and Cloud AISBL and first published in Otto and Tardieu (2025), a policy commentary in Le Grand Continent. Both authors hold or held board positions at Gaia-X AISBL and IDSA. The compilation aggregates EU-level programmes (Digital Europe Programme work programmes 2021-2022 and 2023-2024, EU4Health, Horizon Europe and Destination Earth) with national funding from six member states (Germany, Spain, France, Luxembourg, Denmark and Finland). It includes programmes whose scope extends beyond data spaces as narrowly defined, notably Destination Earth (EUR 240 million) and a broad EU4Health allocation (EUR 280 million). It omits known national investments from other member states. No published methodology accompanies the compilation. The figure should therefore be read as an indicative order of magnitude rather than a precise audited total; the actual aggregate, once additional member state investments are included, is likely higher.
← 6. The workshop was governed by the Chatham House Rule. A summary of workshop discussions is provided in Annex A. For the agenda, see Data Spaces Symposium (2025[26]).