Summary A startup cannot independently apply to be recognised as a frontier AI priority project under the proposed Cloud and AI Development Act (CADA). As proposed, Article 8 explicitly requires that such projects be undertaken by a European digital infrastructure consortium (EDIC) or another legal entity eligible for Union funding, and must involve the participation of at least three Member States. Consequently, a single startup acting alone is structurally ineligible. However, startups can participate as key partners within these consortia. If the consortium is recognised, the startup may benefit from the Union's commitment to match AI computing resources contributed by Member States, as outlined in Article 9.

Detail

The proposed Cloud and AI Development Act (CADA), COM(2026) 502 final, establishes a targeted framework to accelerate the development of frontier AI technologies across the European Union. A cornerstone of this framework is the designation of "frontier AI priority projects," which unlocks access to strategic computing resources and Union-level support. For startups and small-to-medium enterprises (SMEs) operating in the AI sector, understanding the strict eligibility criteria for these projects is critical, as the legislation deliberately channels support through collaborative, cross-border structures rather than individual entities.

The Structural Barrier: Article 8(b) and the Consortium Requirement

The proposal creates a clear distinction between individual innovation and strategic, Union-scale deployment. Article 8 of the CADA proposal sets out the specific criteria the Commission may use to recognise a project as a "frontier AI priority project." The text is unambiguous regarding the legal standing of the applicant.

Crucially, Article 8(b) stipulates that 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 provision establishes a dual requirement that acts as a structural barrier for individual startups:

  1. Legal Entity Requirement: The applicant cannot be a single startup. It must be a "European digital infrastructure consortium" (EDIC) established under Decision (EU) 2022/2481, or "another legal entity eligible for funding under Union law." While a startup might be eligible for funding in other contexts, the specific wording here ties the application to a consortium structure or a specific eligible entity type that implies a broader, multi-partner scope.
  2. Cross-Border Participation: The project must "involve the participation of at least three Member States." This is a quantitative threshold that a single startup, regardless of its size or ambition, cannot meet in isolation. It necessitates a coalition of partners operating across at least three different EU jurisdictions.

Therefore, under the proposal as drafted, a startup acting alone cannot submit an application for recognition. The legislation is designed to ensure that frontier AI projects are not merely national or local initiatives but are integrated into a broader European ecosystem, leveraging the collective capacity of multiple Member States.

How Startups Can Participate: The Consortium Model

While the proposal excludes individual startups from being the primary applicant, it does not exclude them from the benefits. The CADA framework is explicitly designed to foster collaboration between research institutions, public entities, and private industry, including innovative startups. The mechanism for participation is through the consortium model mandated by Article 8.

Startups can participate by:

  • Joining an Existing EDIC: Many European digital infrastructure consortia already exist or are being formed to address specific "grand challenges." Startups can become members of these consortia, bringing their specific technological expertise, agile development capabilities, or niche AI models to the table.
  • Forming a New Consortium: A startup can take the initiative to form a new consortium. In this scenario, the startup would act as a lead or key partner, but it would need to secure partners in at least two other Member States to satisfy the "three Member States" requirement. The resulting legal entity (the consortium) would then be the applicant.

In this model, the startup acts as a key partner or beneficiary within the larger consortium structure. The consortium itself submits the application for recognition to the Commission. This approach aligns with the broader EU strategy of pooling resources and expertise to tackle the "grand challenges" outlined in Annex I of the CADA proposal, specifically Grand Challenge 3, which focuses on developing the next generation of multimodal frontier AI models and systems.

The Benefit: Access to Computing Resources via Article 9

The primary incentive for startups to join these consortia is access to high-performance computing (HPC) resources, which are often prohibitively expensive or unavailable to smaller entities. Article 9 of the CADA proposal addresses the allocation of AI computing resources to recognised projects.

Article 9(1) states that the Union and Member States shall ensure that sufficient AI computing resources are allocated to support the development of frontier AI priority projects that fulfil the criteria set out in Article 8.

The most significant benefit is detailed in Article 9(2), which provides a powerful matching mechanism:

"The Union shall at least match the AI computing resources contributed by Member States to frontier AI priority projects to the extent that sufficient AI computing capacity is available within the Union's share of European high performance computing access time."

This provision creates a multiplier effect for the resources available to the project. If a Member State contributes compute time to a recognised project, the Union is committed to matching that contribution. For a startup involved in the project, this translates to access to substantial computational power that would otherwise be out of reach. This support is critical for training and developing frontier AI models, which require massive computational capabilities and large-scale data processing.

It is important to note that this matching is subject to availability: "to the extent that sufficient AI computing capacity is available within the Union's share of European high performance computing access time." This ensures that the Union's commitment is realistic and tied to the actual capacity of the EuroHPC infrastructure.

The Application Process and Grand Challenges

The recognition process is initiated through "open calls for expressions of interest." Projects are selected based on their alignment with the specific "grand challenges" set out in Annex I of the CADA proposal. Grand Challenge 3 is dedicated to "Frontier AI," focusing on the architectural design and development of next-generation multimodal models and systems that push the boundaries of current algorithmic capabilities.

The Commission assesses whether the project meets the cumulative criteria in Article 8:

  1. It is a pioneering project focused on scaling up frontier AI technologies.
  2. It is undertaken by an EDIC or eligible entity.
  3. It involves the participation of at least three Member States.
  4. The participating Member States pool computing time and other relevant resources.

Startups should monitor these open calls and identify potential partners in other Member States to form a compliant consortium. Early engagement with national authorities and existing EDICs can help startups position themselves as valuable contributors to these large-scale initiatives.

What this means for you

For cloud service providers, data centre operators, and AI startups, the CADA proposal offers a pathway to access strategic resources, but it requires a fundamental shift in how projects are structured and proposed. The era of the "lone wolf" startup applying for Union-level frontier AI status is over; the future is collaborative.

  1. Consortium Building is Non-Negotiable: You cannot apply for frontier AI priority project status alone. You must actively seek partnerships with other entities in at least two other Member States. Look for opportunities to join existing European digital infrastructure consortia (EDICs) or form new alliances with research institutions, universities, or other tech companies across the EU. The legal entity applying must be the consortium, not the startup itself.
  2. Focus on Cross-Border Collaboration: Ensure your project proposal explicitly highlights the participation of at least three Member States. This is a hard requirement under Article 8(b). Document the roles, contributions, and resource pooling of each Member State clearly in your application. The Commission will verify this participation before granting recognition.
  3. Leverage Compute Matching: Understand that the primary value of recognition lies in the compute matching under Article 9. Structure your project's resource requirements to maximise the benefit of this matching. Engage with your national Member State authorities to understand how they can contribute initial compute resources, which the Union will then match. This can significantly reduce the cost of training frontier models.
  4. Monitor Calls for Expressions of Interest: Stay alert for open calls issued by the Commission. These will specify the thematic priorities (e.g., Grand Challenge 3) and application procedures. Prepare your consortium and project documentation well in advance to respond effectively. The selection process is competitive and based on the project's ability to address the grand challenges.

By aligning with these requirements, startups can overcome the structural barriers and access the significant computational support offered by the CADA framework, accelerating their development of frontier AI technologies and contributing to the Union's strategic autonomy.

Common misconceptions

Misconception 1: A startup can apply directly to the Commission. Many startups believe they can submit a project proposal directly to the European Commission for recognition as a frontier AI priority project. This is incorrect. Article 8(b) mandates that the applicant must be an EDIC or an eligible legal entity, and the project must involve at least three Member States. Direct individual applications by startups will not be considered. The application must come from the consortium structure.

Misconception 2: Recognition is automatic for innovative AI projects. Not all AI projects qualify. The project must be selected through an open call for expressions of interest and must support the specific "grand challenge" related to frontier AI (Grand Challenge 3 in Annex I). Furthermore, it must meet all the cumulative criteria in Article 8, including the structural and cross-border participation requirements. Innovation alone is insufficient without the correct legal and geographic structure.

Misconception 3: The Union will provide unlimited compute resources. While Article 9(2) states the Union will "at least match" the resources contributed by Member States, this is subject to the availability of sufficient AI computing capacity within the Union's share of European High Performance Computing (EuroHPC) access time. It is not an unlimited guarantee but a proportional match based on available capacity. The matching is contingent on the Union's ability to provide the resources.

Misconception 4: Only large corporations can form EDICs. EDICs are designed to be inclusive and are not restricted to large corporations. While they require a certain scale of operation to meet the cross-border requirements, they can include startups, SMEs, research institutions, and public entities. The key is the collective structure and cross-border nature, not the size of any single participant. Startups can play a crucial role within these consortia, bringing agility and specialized expertise.

Misconception 5: A startup can apply if it has partners in one other country. The requirement is explicit: "participation of at least three Member States." A project involving only two Member States (e.g., a startup in France and a partner in Germany) would not meet the criteria of Article 8(b). A third Member State must be involved in the consortium and the project to qualify.

Official sources

Related

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