Summary Under the proposed Cloud and AI Development Act (CADA), the phrase "within the limits of available capacity" in Article 9(1) establishes that the Union and Member States are not obligated to guarantee unlimited computing resources for frontier AI projects. Instead, the duty to allocate compute is strictly bounded by the actual, existing physical and virtual capacity within the European High-Performance Computing (EuroHPC) ecosystem. This provision prevents over-commitment, ensuring that resource allocation remains realistic, proportional, and tied to the tangible share of EuroHPC access time available to the Union. The Union's specific "matching" obligation in Article 9(2) is further capped by the availability of capacity within the Union's specific share of EuroHPC access time.

Detail

The Cloud and AI Development Act (CADA), as proposed in COM(2026) 502 final, establishes a structured framework to accelerate the development of frontier AI in the European Union. A critical component of this framework is the allocation of computational resources to projects recognized as "frontier AI priority projects." These projects are designated by the Commission under Article 8, provided they meet strict criteria: they must be pioneering, focused on scaling frontier AI technologies, involve broad participation from at least three Member States, and be undertaken by a European digital infrastructure consortium or another eligible legal entity.

Once a project is recognized, the question of resource allocation arises. Article 9(1) of the CADA proposal sets out the primary obligation:

"The Union and the Member States shall ensure that sufficient AI computing resources from their compute capacities are allocated to support the development of frontier AI priority projects that fulfil the criteria set out in Article 8, within the limits of available capacity."

This qualifying phrase is legally significant. It defines the outer boundary of the Union's and Member States' obligations. It means that the obligation to provide compute is not absolute or infinite; it is conditional upon the existence of spare, usable capacity within the Union's shared infrastructure. The use of "shall ensure" creates a binding duty, but the "within the limits" clause acts as a necessary qualifier that prevents the regulation from creating an impossible legal mandate.

The Binding Nature of the Union's Matching Obligation

While the general allocation duty in Article 9(1) is bounded by available capacity, Article 9(2) introduces a more specific, binding mechanism for the Union's contribution:

"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 "matching" obligation. If Member States contribute a certain amount of compute time to a recognized frontier AI project, the Union is required to contribute an equivalent amount. However, this matching duty is also explicitly capped by the phrase "to the extent that sufficient AI computing capacity is available within the Union's share of European high performance computing access time."

This ties the Union's financial and infrastructural commitment directly to the EuroHPC (European High-Performance Computing) initiative. The Union cannot promise more compute than it actually controls or has reserved within the EuroHPC framework. The "Union's share" refers to the specific allocation of access time that the EU budget and policy have secured within the broader EuroHPC Joint Undertaking structure. If the Union's share is fully booked or technically constrained, the matching obligation cannot be fulfilled beyond that physical limit.

No Guarantee of Over-Commitment

The phrase "within the limits of available capacity" serves as a critical safeguard against over-commitment. In the fast-evolving landscape of AI, demand for compute often outstrips supply. By including this limitation, the CADA proposal acknowledges three key realities:

  1. Physical Constraints Exist: Data centers and supercomputing facilities have finite hardware resources (GPUs, TPUs, memory, bandwidth). The regulation cannot conjure hardware that does not exist.
  2. Shared Resources: EuroHPC resources are shared among various scientific, industrial, and public sector users. Frontier AI projects do not have exclusive, preemptive rights to displace other legitimate users who have already been allocated resources.
  3. Proportionality: The support provided must be proportional to what is technically feasible. The Union and Member States are not required to build new infrastructure instantly to meet a specific project's demand if such capacity does not already exist or cannot be procured within the relevant timeframe.

Interaction with National Strategies and Other Projects

Article 9(3) further clarifies the scope of this allocation by stating that the Union and Member States shall "endeavour to provide sufficient computing resource for AI industrial innovation, physical AI and public sector AI projects." Note the shift in language from "shall ensure" (in Art 9(1)) to "shall endeavour to provide" (in Art 9(3)). This indicates a softer, aspirational obligation for non-frontier AI projects, reinforcing that the hard commitment with capacity limits is primarily focused on the designated frontier AI priority projects.

Furthermore, the allocation of these resources is "without prejudice to the rules and procedures laid down in Council Regulation (EU) 2021/1173" (the EuroHPC Joint Undertaking regulation). This means that the CADA's allocation mechanisms must operate within the existing governance and access policies of the EuroHPC JU. The EuroHPC access policy must be "accommodated to reflect the allocation of such computing resources in an efficient, transparent and timely manner," but it cannot violate the continuity of ongoing operations or the rights of projects already benefiting from allocated EuroHPC AI computing resources.

What this means for you

For in-house counsel, compliance officers, and project leads in AI-driven organizations, understanding the "within the limits of available capacity" clause is crucial for managing expectations and structuring project proposals.

  1. Manage Stakeholder Expectations: Do not represent to investors or partners that recognition as a "frontier AI priority project" guarantees a specific, unlimited amount of compute. The guarantee is conditional on availability. Your legal position must be that the EU will allocate resources if and only if they are available within the EuroHPC share.
  2. Project Feasibility Planning: When designing a frontier AI project, you must account for the possibility that compute allocation may be partial or delayed due to capacity constraints. Your business case should not rely on a guaranteed, full-scale compute provision from the start.
  3. Evidence of Availability: If you are a Member State or a Union entity tasked with allocating resources, you must maintain clear records of "available capacity." This includes documenting the total EuroHPC share, current allocations, and any technical constraints. This documentation is essential to demonstrate compliance with Article 9(1) and (2) and to defend against claims of non-allocation.
  4. Coordination with EuroHPC JU: Ensure that your organization's allocation plans are aligned with the EuroHPC Joint Undertaking's access policies. Any allocation under CADA must be compatible with the existing rights of other EuroHPC users. Over-booking resources could lead to conflicts with the EuroHPC JU's governance rules.
  5. Monitoring Changes in Capacity: The "available capacity" is a dynamic variable. As new data centers come online or as other projects conclude, availability changes. Compliance officers should monitor these changes to ensure that allocation decisions are based on current, accurate data.

Common misconceptions

Misconception 1: Recognition as a frontier AI priority project guarantees full compute funding. Reality: Recognition under Article 8 triggers the obligation to allocate resources, but that obligation is capped by "available capacity" under Article 9(1). If capacity is scarce, the allocation may be limited or delayed.

Misconception 2: The Union must build new infrastructure to meet the demand of frontier AI projects. Reality: Article 9 refers to allocating resources from existing compute capacities. While CADA promotes the deployment of new data centres (Title III), Article 9's allocation duty is not a mandate for immediate, project-specific infrastructure construction. It is an allocation duty based on what is already available or scheduled.

Misconception 3: Frontier AI projects have priority over all other EuroHPC users. Reality: The proposal states that the EuroHPC access policy must be accommodated "without prejudice to the continuity of ongoing operations and the rights of projects already benefiting from allocated EuroHPC AI computing resources." Frontier AI projects do not automatically displace existing users.

Misconception 4: The "matching" obligation in Article 9(2) is unlimited. Reality: The Union's duty to match Member State contributions is explicitly limited to "the extent that sufficient AI computing capacity is available within the Union's share of European high performance computing access time." If the Union's share is full, the matching stops.

Related

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