Summary Under the proposed Cloud and AI Development Act (CADA), frontier AI priority projects are not restricted to using only European High-Performance Computing (EuroHPC) resources. As proposed, Article 9(1) mandates that the Union and Member States allocate sufficient AI computing resources from their broader compute capacities to support these projects. While Article 9(2) specifically ties the Union's matching of Member State contributions to the "Union's share of European high performance computing access time," Member States retain the flexibility to contribute national compute resources outside the EuroHPC framework. This creates a mixed-source model where projects can leverage both Union-backed EuroHPC time and national infrastructure, provided the Union's matching contribution is sourced from the EuroHPC pool.

Detail

The proposed Cloud and AI Development Act (CADA), COM(2026) 502 final, establishes a robust framework to support the development of frontier AI, defined in Article 2(4) as "AI models or AI systems built upon such models that can perform a wide variety of tasks and that approach, reach or exceed the current state of the art." To ensure the Union retains strategic autonomy and reduces dependencies on third-country providers, the proposal introduces "frontier AI priority projects" under Article 8. These projects are selected through open calls and must involve broad participation from entities across the Union, often via European Digital Infrastructure Consortia (EDICs).

The mechanism for providing the necessary computational power to these strategic projects is detailed in Article 9, titled "Computing support for AI projects." This article distinguishes between the general obligation to provide resources and the specific mechanism for Union matching, allowing for a flexible, multi-source approach to compute allocation.

1. Broad Allocation Obligation: Article 9(1)

Article 9(1) establishes the foundational obligation for resource allocation. The text states:

"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."

The phrasing "from their compute capacities" is deliberately broad. It does not limit the source of these resources to the EuroHPC Joint Undertaking (JU) alone. Instead, it encompasses the total pool of AI computing resources available to the Union and its Member States. This implies that Member States may draw from:

  • National supercomputing centers not part of the EuroHPC JU.
  • National "AI factories" or gigafactories.
  • Private-sector partnerships or national cloud infrastructure that meets the project's requirements.

The obligation is to ensure sufficiency of resources, subject only to the constraint of "available capacity." This allows Member States to utilize their full national portfolio to support frontier AI initiatives, ensuring that the Union's strategic goals are met even if EuroHPC capacity is fully utilized or if specific project needs require non-EuroHPC infrastructure.

2. Union Matching Tied to EuroHPC: Article 9(2)

While the allocation obligation is broad, the mechanism for the Union's financial and operational matching is more specific. Article 9(2) states:

"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 critical distinction:

  • The Trigger: The Union's matching obligation is triggered by Member State contributions.
  • The Source: The matching resources must come from the "Union's share of European high performance computing access time."

This explicitly links the Union's contribution to the EuroHPC ecosystem. If a Member State contributes 100 hours of compute from a national non-EuroHPC facility, the Union is obligated to match this contribution, but the matching hours must be drawn from the EuroHPC pool. The matching is conditional on availability ("to the extent that sufficient... capacity is available").

This structure ensures that the Union leverages its specific high-performance computing investments to amplify national efforts, while still allowing Member States to utilize their diverse national assets. It prevents the Union from being forced to match contributions with resources it does not control (i.e., non-EuroHPC national assets), while ensuring that the Union's EuroHPC share is actively deployed for strategic frontier AI projects.

3. The Mixed-Source Model and Member State Flexibility

The interplay between Article 9(1) and Article 9(2) creates a hybrid, mixed-source model for frontier AI projects:

  1. Member State Contribution: A Member State can contribute compute from any of its available capacities (EuroHPC, national supercomputers, private national clouds).
  2. Union Matching: The Union matches this contribution using only its EuroHPC share.
  3. Result: The project receives a combined resource pool: national resources (potentially diverse in architecture and location) plus Union-matched EuroHPC resources.

This flexibility is crucial for projects with specific technical requirements that might not align perfectly with the standard EuroHPC architecture, or for projects where national infrastructure is more readily available. It also respects the sovereignty of Member States to manage their own national compute assets while ensuring a unified Union-level support mechanism via EuroHPC.

Furthermore, Article 9(3) reinforces this broad scope 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." This suggests that the compute support framework is designed to be expansive, covering various AI domains beyond just frontier AI, and relying on the full spectrum of available Union and national capacities.

4. Strategic Context and Sovereignty

The rationale for this mixed-source approach is rooted in the broader objectives of CADA. Recital 34 notes that frontier AI projects are "technical[ly] complex and capital-intensive," requiring a "collaborative approach at Union level." By allowing Member States to contribute diverse national resources, the proposal maximizes the total compute available for these critical projects.

Simultaneously, tying the Union's matching to EuroHPC ensures that the Union's strategic assets are utilized efficiently and remain under EU jurisdiction. This aligns with the sovereignty framework established in Title IV, ensuring that the compute supporting frontier AI is resilient and free from third-country interference. The model effectively balances national flexibility with Union-level strategic coordination.

What this means for you

For CTOs, researchers, and project leaders aiming to secure compute for frontier AI priority projects, understanding the Article 9 distinction is vital for resource planning:

  • Diversify Your National Partnerships: Do not assume you must rely solely on EuroHPC for your national contribution. Engage with your national authorities to identify all available "compute capacities," including national AI factories, private national clouds, or university supercomputers. These can form the base of your project's resource allocation under Article 9(1).
  • Leverage the Matching Mechanism: To maximize your total compute budget, structure your project to include a significant national contribution. This will trigger the Union's matching obligation under Article 9(2), effectively doubling your public-sector compute access (national + EuroHPC). However, be aware that the Union's portion will specifically be EuroHPC time.
  • Plan for Availability Constraints: Both Article 9(1) and 9(2) are subject to "available capacity." If EuroHPC access time is fully booked, the Union's matching contribution may be limited, even if your national contribution is substantial. Early engagement with national competent authorities and the EuroHPC JU is essential to secure slots.
  • Architectural Flexibility: If your project requires specific hardware or low-latency configurations not available on EuroHPC, you can still proceed by sourcing these from national capacities. The Union matching will simply provide the complementary EuroHPC resources, creating a heterogeneous compute environment for your project.

Common misconceptions

  • Misconception 1: Frontier AI projects must use only EuroHPC infrastructure.
    • Reality: Article 9(1) mandates allocation from "compute capacities" broadly. Member States can contribute national resources outside EuroHPC. The restriction applies only to the source of the Union's matching contribution, not the total project resources.
  • Misconception 2: The Union will match national compute contributions with equivalent national resources.
    • Reality: Article 9(2) explicitly limits the Union's matching to the "Union's share of European high performance computing access time." The matching is structurally tied to the EuroHPC pool, not to the specific type of resource contributed by the Member State.
  • Misconception 3: Access to compute is guaranteed regardless of capacity.
    • Reality: The proposal includes clear qualifiers: "within the limits of available capacity" (Article 9(1)) and "to the extent that sufficient... capacity is available" (Article 9(2)). Access is not absolute; it depends on the actual availability of resources in both national and EuroHPC pools.

Related

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