What data spaces actually charge for, what they have failed to charge for, and what the one registry we have says about how many of them work.
Primary source: §1.1, §3.3, §4.2 and Ch. 5 of OECD, Data Spaces between Co-ordination Challenges, Financial Viability and Public Policy, 2026.
The state of the field, in numbers
From the IDSA Radar (dataspaces-radar.org), data as of February 2026
— a voluntary self-reported registry, with the quality caveats below:
- 235 entries, up from 87 in 2022.
- 128 (54 %) at implementation stage.
- 25 (11 %) operational — “approximately one in nine”.
- 0 scaling. (The public dashboard shows one entry classified as scaling; that entry carries no development stage in the dataset export.)
- EU funding is 46.1 % of implementation-stage initiatives and ~20 % of operational ones — “EU funding is associated with reaching implementation but less associated with operational status”.
- Of 96 initiatives reporting an access model, two charge a fee.
- Of 153 implementation- and operational-stage initiatives, only 45 score on all three defining mechanisms.
Independent corroboration on revenue models: only 15 % of European data-sharing initiatives have a clear revenue model (Moonen et al., 2025). And on what the remaining ones think they are: of 20 initiatives studied, 8 framed themselves as data marketplaces, though orchestrators admitted in interviews that the actual goal had been deeper collaboration (Eustache, Brousseau & Toledano, 2025).
How much the numbers can carry
Box 1.1 of the report is unusually explicit, and the caveats should travel with every figure above: registration is voluntary and European-biased; ~50 % of entries lack geographical focus, 58 % lack funding source, 78 % lack start dates; failed initiatives may remain listed as operational (Agdatahub’s bankruptcy is not reflected); the Radar is operated by IDSA, so non-IDSA stacks may simply not register. The OECD promotes this to a finding rather than burying it: “This measurement gap is itself a key finding.” Its structural explanation — “no single entity observes all transactions” — is a direct consequence of federation: “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.”
The marketplace model does not work
Three independent signals, all pointing the same way:
- Adoption data. The marketplace building block appears in 18 % of all Radar initiatives but only 8 % of operational ones — it drops out as initiatives mature.
- Pricing data. Two of 96 initiatives charge for access.
- Self-report. Orchestrators who described their initiative as a marketplace said the real objective was collaboration (Eustache, Brousseau & Toledano, 2025).
The theoretical reason is the one from §3.1: price formation for data fails. Value is context-dependent; Arrow’s information paradox means a buyer cannot assess information without receiving it, so “price negotiation requires revealing the very asset being sold”; and in multi-party settings the same data are worth different amounts to each party, so no single price clears.
Meanwhile the provenance building block rises from 25 % of all initiatives to 48 % of operational ones, and access-and-usage policy falls slightly (35 % → 28 %). ⟨Inference: taken together these are a portrait of what survives contact with production — not a market for data but an audit trail for data that was going to move anyway.⟩
What viable ones charge for instead
The report’s policy consideration 4: support business models built around infrastructural value — cost sharing and membership fees for the commons layer, with revenue from value-added services on top, rather than from data monetisation.
- ABtC: member fees plus transaction fees; collaborative (non-competitive) services first, competitive services layered on later.
- Catena-X: association membership funds the commons; the operating environment is a separate layer where commercial operators compete.
- T-Systems (Box 3.5): the commercial-operator model, and the only revenue projection in the report — its Data Intelligence Hub provides Catena-X onboarding, took the first IDSA connector certification in January 2024, built a Japanese test environment with NTT in September 2024, and co-founded the IDSA Global Telecom Community (June 2024, with NTT Communications and KPN). NTT DATA targets roughly EUR 310 million in data-space-related revenue by 2030. ⟨Inference: the revenue being projected is integration and operations services, not data — the commercial bet is on the picks-and-shovels layer.⟩
- Public utilities: X-Road and the EU Public Procurement Data Space do not have a revenue model and are not supposed to. Where value “cannot be captured through market-based mechanisms”, sustainability means recurrent budget — a permanent utility. X-Road’s two decades and >1 bn transactions/year are the proof that this is a real model rather than a concession.
The funding critique — the instruments were built for another job
The sharpest institutional finding in the report, and it indicts the funder rather than the funded. Two distinct purposes are routinely conflated:
| Purpose | Rationale | What success looks like |
|---|---|---|
| Fund the technical foundation | Pre-competitive R&D with spillovers: schemas, protocols, interoperability frameworks | A standard exists and is adopted |
| Fund ecosystem creation | A specific collective-action problem no market actor has the incentive or legitimacy to solve | An ecosystem operates without the grant |
“EU funding programmes were not explicitly designed around this distinction or the bootstrapping rationale… The bootstrapping justification is an analytical overlay by economists, not the stated rationale of the policymakers who designed the funding instruments.”
The stated rationales were competitiveness, digital sovereignty and the single market for data. Hence:
“The instruments were simply not designed for the problem they are now expected, or may need, to solve.”
Two reinforcing mechanisms:
- Multitask principal-agent misalignment (Holmström & Milgrom, 1991; Lerner, 2009 on public venture funding): where funding rewards observable compliance — consortium formation, architectural alignment, dissemination — over unobservable operational viability, rational actors optimise for the observable. “Compliance is measurable; progress towards operational viability is not.”
- Dependency. From the Expert Workshop: continued public support “may remove the pressure to find viable revenue models altogether”. The Dutch CoE-DSC’s own figure — nearly 80 % of its 47 catalogued initiatives depend on government funding — is the scale of the exposure.
And the standing caution that should be quoted whenever “public infrastructure” is offered as a justification:
“That many data spaces depend on public funding does not in itself demonstrate that they are public infrastructure. It may instead indicate that viable business models have not yet been found.”
The EUR 1.7 billion figure, handled honestly
EUR 1 735.4 million of EU public investment as of early 2025 — but the report’s endnote on it is a model of provenance discipline and should travel with the number. It originates in a Gaia-X AISBL funding portfolio first published by Otto and Tardieu (2025) in Le Grand Continent; “both authors hold or held board positions at Gaia-X AISBL and IDSA”; it aggregates DEP 2021-22 and 2023-24, EU4Health, Horizon Europe, Destination Earth (EUR 240 m) and an EU4Health allocation (EUR 280 m) with national funding from six member states; it includes programmes beyond data spaces narrowly defined, omits other member states, and has no published methodology. The OECD reads it as “an indicative order of magnitude rather than a precise audited total”, probably on the low side.
The demand side is where almost nobody spends
Policy consideration 3. Funding has gone overwhelmingly to supply — building infrastructure — while adoption is limited by the cost and capability of the firms expected to connect. Three counterexamples, and they are rare enough that the report can name them all:
- Spain, Data Spaces Kit — Red.es, EUR 60 million from NextGenerationEU; non-competitive grants up to EUR 30 000 per private data provider and EUR 50 000 per public administration. Explicitly “breadth of adoption rather than depth of capability”.
- Japan, IPA Data Spaces Academy — guides, a CDO handbook, a data maturity framework; attacks the informational barrier rather than the financial one.
- India, DEPA — regulatory design that creates a commercially viable intermediary category (RBI-licensed Account Aggregators, >270 million linked accounts) instead of subsidising infrastructure. The report’s own comment: “creating a commercially viable intermediary category may address the bootstrapping problem more durably than subsidising infrastructure.”
AI as the demand event
Policy consideration 7, and the report’s one genuinely forward-leaning claim: AI-driven demand for governed, traceable, high-quality training and inference data is “an economic necessity rather than a policy aspiration” for data-space viability — conditional on the stack delivering governed access “at the speed and scale that AI development requires”. The EU’s Data Union Strategy (Nov 2025) explicitly repositions data spaces from a governance instrument to AI supply-chain infrastructure, and introduces “data labs”.
The commercial comparators make the speed requirement concrete: Snowflake reported USD 4.7 bn revenue in FY2026 and Databricks USD 5.4 bn annualised as of January 2026 — over USD 10 bn combined across more than 30 000 customers, with roughly a quarter of Snowflake customers maintaining sustained cross-account sharing. These meet the broad definition of a data space and not the narrow one; Databricks open-sourcing Delta Sharing and Unity Catalog narrows the interoperability gap without conceding federated governance or provider-retained usage control.
⟨Inference: the uncomfortable reading is that the commercial stack has demonstrated revenue at a scale the governed stack has not demonstrated at all, and is closing on it from the interoperability side. The governed stack’s remaining defensible ground is exactly the two properties the boundary conditions describe — neutrality where participants compete, and control that survives the handover.⟩
The unanswered IP question
Flagged twice in the report and left open: usage control governs how data may be processed but does not resolve who owns the value derived from it — model ownership, derived insights, downstream patent claims. The worked case is the US Materials Genome Initiative: over USD 1 billion cumulative federal investment and persistent industry reluctance to contribute proprietary performance and testing data, over IP attribution. PETs, federated learning and revenue-sharing coalitions address parts of it; “none has yet produced a widely adopted commercial model”.
⟨Inference: this is distinct from the confidentiality gap. That gap asks can my input be seen; this asks who owns what was learnt from it. Cryptography answers the first and is silent on the second, which is a limit worth stating explicitly wherever we argue the MPC case. The wiki has no page on derived-value ownership; it is a candidate.⟩
Three field instances from plastics (POLYPROBLEM 06, 2023) ⟨2026-11-06⟩
Source: Der Circularity Code. Everything above comes from the OECD’s survey of data spaces. These are the same questions answered by operators who did not know they were building data infrastructure.
1. Commons vs. interfaces, argued by two operators — and the financing objection that does not discriminate
Reuse systems for take-away packaging, pp. 24–25. Markus Urff (Stiftung ReFrastructure) wants a commons: “Wir betrachten eine künftige Mehrweg-Infrastruktur als ein gesellschaftliches Gemeingut”, with the analogy that BMW, Mercedes and VW do not build their own roads — “Bei der digitalen Mehrweginfrastruktur handelt es sich letzten Endes um eine digitale Datenautobahn.” Fabian Barthel (Vytal) refuses it: “Ich halte ein solches gemeinsames System für nicht notwendig” — return machines that read any provider’s container IDs, “durchaus Schnittstellen zwischen den Systemen, aber eben keine gemeinsame Daten-Plattform”. The report frames the split as Gemeinwohlökonomie vs. competitive differentiation.
The useful part is what they share (p. 25):
Weder für eine gemeinnützige digitale Infrastruktur noch für ein Netz von Rücknahmestationen im öffentlichen Raum und die dahinter liegende Logistik existiert ein tragfähiges Geschäftsmodell. Niemand verdient mit der Rücknahme. Sie verursacht nur Kosten.
⟨Inference: this is the cleanest documented instance in the wiki of the federation-vs-interoperability choice being made by operators rather than architects, and it cuts against the usual framing. Interoperability-only is often presented as the cheaper path; here it has the identical revenue problem. What it is cheaper at is agreement, not operation — which is a real advantage, but a different one. The OECD’s four boundary conditions (data-space boundary conditions) test whether a federation beats bilateral contracts; neither party here passes the viability test at all.⟩ The instrument proposed is an earmarked single-use levy funding the return and washing logistics.
2. Data as the revenue line when the physical service has none
Bin-e’s smart bins, p. 34: they had no financial value to whoever installed them, so Jakub Luboński constructed one out of the data — real-time usage and composition readings in a cloud, read via app or dashboard, because
So lassen sich beispielsweise Stoffströme besser kontrollieren oder aber auch wichtige Daten für das CSR- oder EPR-Reporting generieren.
⟨Inference: a textbook case of the pattern recorded above under what viable ones charge for instead — the willingness to pay is created by a reporting obligation, not by the physical service. It is also the smallest instance in the wiki of compliance demand retrofitted onto hardware that could not pay for itself.⟩
3. Verification as the business model
KOLEKT (Thierry Sanders, Circular Action B.V.), pp. 42–44, combines three functions — marketplace, monitoring and proof of final disposal, and incentives payable to unbanked collectors — and states its revenue thesis as audit:
Das sind auch die Daten, die die Wirtschaftsprüfer von Unternehmen sehen wollen und letztlich auch für die EPR-Berichterstattung am wichtigsten sind. […] unser Ziel ist es, uns zu einer Auditierungsplattform für Abfälle zu entwickeln.
Joel Tasche (CleanHub), p. 40, states the cost side of the same business: “Standards sind wenig wert, wenn sie nicht eingehalten werden. Gleichzeitig kostet die fortlaufende Dokumentation Zeit und somit auch viel Geld, das dem Abfall- und Recyclingsektor vielerorts fehlt” — with ML detecting double-booking and fraud, but social audits still needing to be physical and on site.
⟨Inference: both firms monetise attestation, not data access. That is a fourth answer to “what do viable data spaces charge for” alongside the ones recorded above, and the one with the clearest payer: whoever bears a reporting liability. It is the same conclusion residual-value grading reaches from the Japanese side — verification is paid for by its beneficiary.⟩
The standards-proliferation cost, measured
Ellen Gunsilius and Steffen Blume (GIZ), p. 41, count more than 20 national and international waste-collection apps in Indonesia alone and derive a need for “Harmonisierung und Interoperabilität”, because otherwise “weiterer Fortschritt ausgebremst und Ressourcen für die immer gleichen Ansätze investiert würden” — the IDSA-style fragmentation finding above, in a market with no standards body at all.
Cross-references
- OECD, Data Spaces between Co-ordination Challenges, Financial Viability and Public Policy (2026)
- Data space boundary conditions · The bootstrapping problem
- Data Commons — the governance layer these business models are trying to fund
- Coordination failure vs. bargaining failure
- Catena-X · IDSA · Gaia-X
- AI stack market structure — Snowflake/Databricks as the commercial comparator
- Confidentiality gap · Cryptographic trust infrastructure
Sources
sources/research/2026-10-27-oecd-data-spaces-co-ordination-viability-public-policy/2026-10-27-oecd-data-spaces-co-ordination-viability-public-policy.md
(§1.1, Box 1.1, §2.4, §3.1, §3.3, §3.4, §4.2, Boxes 3.4, 3.5, 4.1, Ch. 5).