Summary Under the proposed Cloud and AI Development Act (CADA), the European Commission would recognise a "frontier AI priority project" by means of a formal decision under Article 8. Recognition would be discretionary, not automatic: the provision says the Commission "may" recognise such projects. The route in is selection through open calls for expression of interest, and the project must support grand challenge 3 set out in Annex I. To qualify, a project would have to fulfil all three cumulative criteria in Article 8: (a) it is a pioneering project focused on the support and scaling-up of frontier AI technologies; (b) it is undertaken by a European Digital Infrastructure Consortium (EDIC) established pursuant to Decision (EU) 2022/2481, or another legal entity eligible for funding under Union law, and involves at least three Member States; and (c) the participating Member States pool computing time and other relevant resources. The proposal does not specify how many calls there would be, a cap on recognised projects, a timeline, or a dedicated appeal route.

Detail

CADA — published as COM(2026) 502 final and still only a proposal — introduces a mechanism to identify and support a small set of high-impact artificial intelligence initiatives at Union level. The label for these initiatives is "frontier AI priority project," and the procedure for conferring that label is set out in Article 8 of the proposed Regulation.

The legal basis for recognition

Article 8 provides that "The Commission may, by means of a decision, recognise as frontier AI priority projects, projects selected through open calls for expression of interest that support grand challenge 3 set out in Annex I, provided that the following criteria are fulfilled." Three features of this wording would be decisive in practice.

First, recognition would be discretionary. The operative verb is "may," not "shall." Even where every criterion is satisfied, the Commission would retain a margin of appreciation over whether to recognise a given project. Meeting the criteria would be a necessary condition, not a sufficient one.

Second, recognition would take the form of a Commission decision — a legal act. It would not be a self-declaration by a project lead, nor a finding by a national authority. The proposal, as drafted, does not assign national competent authorities any role in the recognition step itself, and it does not establish an appeal mechanism specific to this decision. (General avenues of review under Union law would, of course, apply to any Commission act; the proposal simply does not create a bespoke one.)

Third, the gateway is the open call for expression of interest. Projects would not apply at will through a standing portal; they would have to be selected through a call published by the Commission. The proposal does not specify how many such calls would be issued, on what schedule, or whether the number of recognised projects would be capped — so those points should be treated as not yet specified rather than assumed.

Alignment with grand challenge 3

Article 8 ties recognition to a single strategic objective: the project must "support grand challenge 3 set out in Annex I." Grand challenge 3 in Annex I is "Developing the next generation of multimodal frontier AI models and systems and pioneering novel capabilities."

The link is narrowing. A project that supports a different grand challenge — or that pursues AI work falling outside the development of next-generation multimodal frontier models and novel capabilities — would not be eligible for frontier AI priority project status under Article 8, however meritorious it might be on its own terms. Counsel should map a proposal to the verbatim wording of grand challenge 3 rather than to a general notion of "AI innovation."

Recital 34 supplies the rationale, in substance: the criteria are framed as they are because frontier AI requires resources at an unprecedented scale, which in turn calls for a collaborative Union-level approach with broad participation — in particular through EDICs established pursuant to Decision (EU) 2022/2481 or any other legal structure representing a meaningful share of the Union's interest, notably in cybersecurity.

The three cumulative criteria

For the Commission to be in a position to recognise a project, all three criteria in Article 8 — (a), (b) and (c) — must be fulfilled. They are cumulative: failure on any one would put recognition out of reach regardless of how strongly the others are met.

(a) Pioneering focus on scaling. The project must be "a pioneering project, focused on the support and scaling-up of frontier AI technologies." The emphasis falls not only on novelty but on scaling-up — moving frontier technologies beyond proof of concept toward operational capability. The provision does not attach any numeric performance or compute threshold to this criterion, and none should be read in.

(b) Eligible legal structure and at least three Member States. The project must be "undertaken by a European digital infrastructure consortium established pursuant Decision (EU) 2022/2481 or another legal entity eligible for funding under Union law and it involves the participation of at least three Member States." This criterion has two limbs that must both hold:

  • Vehicle. The undertaking entity must be either an EDIC under Decision (EU) 2022/2481 — the instrument that created European Digital Infrastructure Consortia under the Digital Decade Policy Programme — or another legal entity eligible for funding under Union law. The EDIC is highlighted in Recital 34 as the paradigm vehicle, but the text expressly leaves room for alternatives.
  • Membership. The project must involve at least three Member States. A purely national project, or a bilateral one between two Member States, would fall short of the threshold.

(c) Resource pooling. The "participating Member States pool computing time and other relevant resources to support the implementation of the designated project." Endorsement alone would not suffice; participating Member States must actually pool computing time and other relevant resources behind the designated project. This pooling is what gives the subsequent support mechanism something to build on.

What recognition unlocks (Article 9)

Article 8 governs recognition; Article 9 governs the support that recognition opens up. As proposed, the Union and Member States would ensure sufficient AI computing resources for recognised projects within available capacity, and the Union "shall at least match" the AI computing resources contributed by Member States — to the extent that sufficient capacity is available within the Union's share of EuroHPC (European high performance computing) access time. The matching commitment is therefore conditional on available capacity rather than absolute. Recognition under Article 8 is best understood as the gateway to this matched-computing arrangement, not as a guarantee of any particular volume of resources.

What this means for you

For in-house counsel and advisers to technology firms, research institutions, or public bodies, Article 8 (as proposed) carries several practical implications. Each is conditional on the proposal being adopted in its current form.

Watch for open calls. Because the entry point is the open call for expression of interest, there would be no standing application route. Monitor Commission announcements; the proposal does not fix a schedule or number of calls, so timing should not be assumed.

Build the consortium and the vehicle early. Criterion (b) means a single company or single national lab cannot qualify alone. Form a cross-border structure involving at least three Member States, and confirm that the chosen vehicle is either an EDIC under Decision (EU) 2022/2481 or another entity eligible for funding under Union law.

Document the resource pooling. Criterion (c) requires demonstrable pooling of computing time and other relevant resources by participating Member States. Secure and document those commitments in a verifiable form.

Map the project to grand challenge 3. Tie the proposal expressly to the development of next-generation multimodal frontier AI models and systems and to pioneering novel capabilities, using the Annex I wording rather than generic language.

Calibrate expectations on discretion. Because recognition is discretionary and made by Commission decision, satisfying the criteria would not create an entitlement to recognition. Prepare the file on that basis.

Common misconceptions

Misconception 1: Any large AI project can be recognised. Reality: Only projects that support grand challenge 3 are eligible. A project aligned to a different challenge, or outside the scope of grand challenge 3, does not qualify under Article 8.

Misconception 2: A single Member State can sponsor a priority project. Reality: Article 8(b) requires the participation of at least three Member States. A national or bilateral project does not meet the threshold.

Misconception 3: Recognition is automatic upon submission. Reality: Recognition is discretionary — the Commission "may" recognise — and is made by formal decision following selection through open calls. Submitting a proposal confers no right to recognition.

Misconception 4: Only EDICs can be the vehicle. Reality: The EDIC is the paradigm vehicle, but Article 8(b) also permits "another legal entity eligible for funding under Union law." The at-least-three-Member-State requirement still applies either way.

Misconception 5: Recognition guarantees a fixed amount of computing. Reality: Article 9 frames the Union's matching of Member State contributions as conditional on sufficient capacity being available within the Union's share of EuroHPC access time. Recognition opens the door to matched computing; it does not guarantee a set volume, and the proposal does not link recognition to any direct financial grant.

Official sources

Related

This is general information about a draft EU regulation, not legal advice.