Summary Under the proposed Cloud and AI Development Act (CADA), the "scale" required for a project to qualify as a frontier AI priority project is not defined by a specific numerical threshold of compute power (such as FLOPs) or parameter counts. Instead, the legislation defines scale structurally and collaboratively. Article 8 mandates that qualifying projects must be "pioneering," address Grand Challenge 3 (frontier AI), and critically, involve the participation of at least three Member States who must pool computing time and other relevant resources. Recital 34 explicitly grounds this requirement in the "unprecedented scale of resources" needed for frontier AI, noting that the "technical complexity and capital-intensive nature" of such projects necessitates a collaborative Union-level approach rather than isolated national or private efforts.

Detail

The proposed CADA (COM(2026) 502 final) establishes a distinct mechanism to identify and support high-impact artificial intelligence initiatives. To understand the "scale" required under this framework, one must look beyond technical metrics and examine the structural and collaborative criteria embedded in the legislation. The Act treats "scale" not as a static measurement of hardware, but as a dynamic requirement for cross-border resource aggregation.

1. The Legal Definition of Scale: Article 8

Article 8, titled "Criteria for frontier AI priority projects," provides the definitive test for scale. The Commission may recognize a project as a frontier AI priority project only if it meets three cumulative criteria, which collectively operationalize the concept of "scale":

  1. Pioneering Nature: The project must be "pioneering, focused on the support and scaling-up of frontier AI technologies." This implies that the scale is not merely about size, but about pushing the boundaries of the current state of the art. The project must align with Grand Challenge 3 set out in Annex I, which focuses on developing next-generation multimodal models and systems.
  2. Legal Structure and Multi-State Participation: The project must be 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. Crucially, it must "involve the participation of at least three Member States." This is the primary proxy for scale in the legislation: the Act assumes that only a multi-state coalition can muster the necessary resources to tackle frontier AI.
  3. Resource Pooling: The participating Member States must "pool computing time and other relevant resources to support the implementation of the designated project."

2. The Rationale: Recital 34 and Resource Intensity

Recital 34 provides the legislative intent behind these strict structural requirements. It explicitly states: "Given the unprecedented scale of resources required for frontier AI development, it is necessary to set criteria for the designation of a project as a frontier AI priority project."

The recital identifies two key drivers that necessitate this definition of scale:

  • Technical Complexity and Capital Intensity: The text notes that "the projects require a collaborative approach at Union level" due to their "technical complexity and capital-intensive nature." The proposal posits that the financial and technical burden of developing frontier AI is too great for a single Member State or a single private entity to sustain alone.
  • Strategic Asset Development: The focus is on developing frontier AI models and systems as "strategic assets," with a specific mention of key sectors such as cybersecurity.

By linking the definition of a priority project to the "unprecedented scale of resources," CADA effectively argues that scale is a barrier to entry that can only be overcome through collective action. The "scale" is therefore legally operationalized as the ability to mobilize a coalition of at least three Member States and pool their sovereign compute resources.

3. Operationalizing Scale: Article 9 and Compute Allocation

The scale required is further clarified in Article 9, "Computing support for AI projects." This article outlines how the Union and Member States must respond to these scaled projects, creating a direct link between the definition of scale and the allocation of public resources:

  • Proportional Matching: The Union shall "at least match the AI computing resources contributed by Member States to frontier AI priority projects," provided sufficient capacity is available within the Union's share of European High Performance Computing (EuroHPC) access time.
  • Sufficient Allocation: Both the Union and Member States must ensure that "sufficient AI computing resources" are allocated to support the development of these projects.

This creates a feedback loop: the "scale" of the project (defined by the multi-state pooling in Article 8) determines the scale of the public compute subsidy. If a project does not meet the multi-state, pooled-resource criteria of Article 8, it does not trigger the matching compute commitments of Article 9. The legislation thus incentivizes the formation of large, cross-border consortia to access the necessary infrastructure.

4. Distinction from the EU AI Act

It is critical to distinguish the CADA's definition of "frontier AI" scale from the EU AI Act's definition of "general-purpose AI models with systemic risk." The AI Act (Regulation (EU) 2024/1689) uses a quantitative threshold (currently 10^25 floating-point operations) to presume systemic risk. The CADA, however, does not cite a specific FLOP count in Article 8.

Instead, CADA uses a qualitative and structural definition. A project could technically use less compute than the AI Act's threshold but still qualify as a "frontier AI priority project" under CADA if it is pioneering, capital-intensive, and involves three Member States pooling resources. Conversely, a model exceeding the AI Act's threshold might not qualify for CADA priority status if it is developed by a single entity without multi-state collaboration. The CADA focuses on the governance and resource aggregation required to build the infrastructure, whereas the AI Act focuses on the risk profile of the resulting model.

What this means for you

For CTOs, architects, research leads, and SMEs, the definition of scale under CADA has significant strategic implications for funding, partnership, and infrastructure planning.

1. Collaboration is Mandatory for Priority Status You cannot achieve "frontier AI priority project" status as a standalone entity, even if your technical capabilities are world-class. The law requires a consortium involving at least three Member States. If you are an SME or a mid-cap provider, your path to accessing the specific compute matching funds under Article 9 is through joining a broader European Digital Infrastructure Consortium (EDIC). You must structure your projects to be inherently collaborative from the outset, identifying partners in at least two other Member States willing to commit resources.

2. Compute Pooling is a Hard Requirement The requirement to "pool computing time" means that your project architecture must be designed to integrate with EuroHPC resources and national compute capacities. You cannot simply claim you need scale; you must demonstrate a concrete plan for how Member States are contributing and pooling their specific compute assets. Your technical architecture must be compatible with these distributed, pooled environments, likely requiring interoperability with national high-performance computing grids.

3. Focus on "Pioneering" and "Strategic" Use Cases The scale is not just about raw power; it is about strategic impact. Recital 34 highlights cybersecurity as a key sector. If your project is simply scaling up an existing commercial model for generic purposes, it may not meet the "pioneering" criterion. Your use case must demonstrate a clear advancement of the state of the art and address a strategic Union interest, such as the development of frontier AI models for cybersecurity or other critical domains identified in Annex I.

4. SMEs and Subcontracting While the primary applicants must be large consortia involving multiple states, SMEs can play a crucial role as specialized partners within these EDICs. The CADA's broader ecosystem approach (Title II) supports the development of open cloud stacks and industrial AI. SMEs should position themselves as providers of specialized components (e.g., specific AI agents, physical AI models, or energy-efficient data center technologies) that feed into these large-scale frontier projects, acting as the technical engine within the multi-state framework.

Common misconceptions

Misconception 1: "Scale" means a specific FLOP count. Many assume that because the AI Act uses a 10^25 FLOP threshold for systemic risk, the CADA does the same for frontier AI. This is incorrect. Article 8 does not specify a compute threshold. It specifies a collaborative threshold (three Member States). A project with lower compute usage but high strategic value and multi-state pooling could qualify, while a massive commercial project developed by a single third-country entity would not.

Misconception 2: Any large AI project qualifies. The term "frontier AI" is defined in Article 2(4) as models that "approach, reach or exceed the current state of the art." However, not all state-of-the-art models qualify as priority projects. They must be "pioneering" and involve the specific legal structure of an EDIC or equivalent. Commercial products developed by private entities without multi-state resource pooling do not meet the Article 8 criteria, regardless of their technical scale.

Misconception 3: The Union will provide all the compute. Article 9 states the Union will "match" the resources contributed by Member States. It does not say the Union will provide the base layer unilaterally. The "scale" required includes a commitment from Member States to contribute and pool their own resources first. The Union's role is to amplify this contribution, not replace it. The project must be driven by the collective will of at least three Member States.

Official sources

Related

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