Summary Under the proposed Cloud and AI Development Act (CADA), public-sector buyers seeking high-performance computing (HPC) for AI projects face a distinct legal regime compared to "frontier AI priority projects." While Article 9(1) and 9(2) create a binding obligation for the Union and Member States to ensure and match resources for designated frontier projects, Article 9(3) establishes a softer "endeavour" obligation for public-sector AI projects. This means public bodies can access compute resources, but they do so on a best-effort basis rather than as a guaranteed right. Crucially, securing this support requires alignment with Article 7 national cloud and AI strategies, which serve as the primary roadmap for resource allocation and strategic prioritization within each Member State.

Detail

The proposed Cloud and AI Development Act (CADA), COM(2026) 502 final, addresses a critical bottleneck in the EU's digital transformation: the scarcity of high-performance computing capacity required to train and deploy advanced AI models. To manage this scarcity, the proposal establishes a tiered allocation framework in Article 9, titled "Computing support for AI projects." This framework differentiates between projects of strategic Union-wide priority and those serving broader industrial, physical, or public-sector needs.

The Hierarchy of Compute Allocation in Article 9

Article 9 creates a clear hierarchy in how the Union and Member States must manage their available compute capacities, primarily drawn from the European High Performance Computing Joint Undertaking (EuroHPC) and national capacities.

1. The Mandatory Mandate for Frontier AI (Article 9(1) & 9(2)) The highest tier of support is reserved for "frontier AI priority projects." These are pioneering projects designated by the Commission under Article 8, which must meet strict criteria, including being undertaken by a European digital infrastructure consortium and involving at least three Member States.

  • Article 9(1) states that the Union and Member States "shall ensure that sufficient AI computing resources... are allocated" to these designated projects.
  • Article 9(2) adds a matching mechanism: "The Union shall at least match the AI computing resources contributed by Member States to frontier AI priority projects" within the limits of available EuroHPC capacity. This creates a "hard" obligation. If a project is designated as a frontier priority, the Union is legally bound to provide matching resources.

2. The "Endeavour" Obligation for Public-Sector AI (Article 9(3)) Public-sector buyers fall into a different category. Article 9(3) explicitly addresses the needs of the public sector, alongside industrial and physical AI. It states:

"The Union and the Member States shall endeavour to provide sufficient computing resource for AI industrial innovation, physical AI and public sector AI projects."

The use of the term "shall endeavour" is legally significant. In EU legislative drafting, this phrasing indicates a strong policy commitment and a duty to make every reasonable effort, but it stops short of creating an absolute, enforceable right to a specific quantity of resources. Unlike the "shall ensure" and "shall match" language in paragraphs 1 and 2, Article 9(3) acknowledges that while public-sector AI is a priority, it competes for finite resources against industrial and physical AI initiatives, and its allocation is subject to availability and feasibility.

Defining "Public Sector AI Projects"

While Article 9 does not provide a standalone definition of "public sector AI projects," the context of the Regulation and its recitals clarifies their scope. These projects involve Union entities or public-sector bodies developing, testing, or deploying AI models and systems to serve public interests.

Recital 22 of the explanatory memorandum elaborates that the Cloud and AI Leadership Initiatives aim to "increase the development and adoption of AI models and systems across the Union's public sector." The text highlights specific use cases:

  • Supporting better decision-making.
  • Simplifying administrative procedures.
  • Reducing unnecessary burdens in critical domains such as healthcare.
  • Facilitating the sharing and reuse of training data and AI models across public services.

Therefore, a "public sector AI project" under CADA could range from training a model to optimize national energy grids to developing diagnostic tools for public hospitals or streamlining border control processes. The key differentiator is the public-sector nature of the entity and the public-interest objective of the project.

The Critical Link to National Strategies (Article 7)

Access to compute resources under Article 9(3) is not an isolated entitlement; it is inextricably linked to national planning. Article 7 mandates that Member States establish national cloud and AI strategies within one year of the Regulation's entry into force.

These strategies are the operational blueprint for how Member States will fulfill their "endeavour" obligations under Article 9. Specifically:

  • Article 7(2)(c) requires national strategies to include "measures to support the broad deployment and uptake of AI in strategic industrial and public sectors."
  • Article 7(2)(e) mandates measures to "invest in high-intensity computing infrastructure, including AI factories, AI gigafactories and quantum computers as strategic national and cross-border assets."

For a public-sector buyer, this means that the likelihood of securing compute resources depends heavily on the alignment of their project with their Member State's national strategy. A project that is not reflected in or supported by the national strategy may struggle to access the "endeavour" pool of resources. The national strategy effectively acts as the filter through which the Union's "endeavour" obligation is translated into concrete national actions.

Integration with the Cloud and AI Leadership Initiatives

The compute allocation in Article 9 is the practical engine for the broader Cloud and AI Leadership Initiatives established in Article 3. These initiatives aim to promote research, innovation, and large-scale capacity.

Operational Objective 7 (Article 4(7)) specifically targets the public sector, aiming to:

  • Accelerate the technological development and uptake of AI models in critical public domains.
  • Develop AI models that improve public service delivery and accessibility.
  • Promote the sharing and reuse of training data and AI models across the Union's public services.

Article 9(3) is the mechanism that enables Operational Objective 7. Without the "endeavour" to provide compute, the public sector could not realistically develop the advanced models envisioned in the Leadership Initiatives. However, the reliance on "endeavour" rather than "guarantee" reflects the Commission's recognition that public-sector projects, while vital, may not always carry the same strategic weight as frontier AI projects that define the Union's global technological leadership.

What this means for you

For public-sector procurement officers, IT strategists, and project managers, the distinction in Article 9 has profound implications for project planning and resource acquisition.

1. Manage Expectations: "Endeavour" is Not "Guarantee"

Do not assume that a public-sector AI project automatically qualifies for a specific allocation of HPC resources. Unlike frontier priority projects, you do not have a statutory right to a matching grant of compute. You must treat access to resources as a competitive process where your project must demonstrate high value to secure the "endeavour" of the Union and Member States.

2. Strategic Alignment is Non-Negotiable

Before applying for compute support, verify that your project is explicitly referenced in your Member State's national cloud and AI strategy (Article 7). If your project is not aligned with the national roadmap, it is unlikely to be prioritized.

  • Action: Review your national strategy document. If your project type is not mentioned, engage with the national competent authority to advocate for its inclusion or alignment with existing strategic priorities (e.g., healthcare, energy, or border management).

3. Leverage the "Public Interest" Argument

Since Article 9(3) groups public-sector AI with industrial and physical AI, you must articulate why your project serves a unique public interest that justifies the allocation of scarce resources.

  • Action: In project proposals, explicitly link your compute needs to the objectives in Article 4(7): improving public service delivery, simplifying administration, or addressing critical domains like healthcare. Highlight how your project facilitates the sharing and reuse of data across the public sector, a key goal of the Leadership Initiatives.

4. Plan for Contingency and Collaboration

Given the "endeavour" basis, resource availability may fluctuate.

  • Action: Develop contingency plans. Consider partnerships with industrial entities or research institutions. Joint projects may have stronger claims on resources, as they bridge the gap between public interest and industrial innovation.
  • Action: Explore the EuroCloud Federation (Article 34) as an alternative or supplementary source of capacity, allowing public bodies to share idle resources among themselves.

5. Engage Early with National Authorities

The "endeavour" obligation is implemented by Member States. Early engagement with national competent authorities is essential to understand the specific criteria they use to prioritize projects under Article 9(3).

  • Action: Initiate dialogue with your national authority during the planning phase, not just when resources are needed. Understand the timeline for the national strategy updates (Article 7(5)) and ensure your project timeline aligns with these cycles.

Common misconceptions

Misconception 1: Public-sector AI projects have the same compute rights as frontier AI projects. Correction: No. Frontier AI priority projects (Article 9(1) & 9(2)) have a binding "shall ensure" and "shall match" obligation. Public-sector projects (Article 9(3)) are subject to a "shall endeavour" obligation, meaning access is based on best efforts and availability, not a guaranteed right.

Misconception 2: National strategies are merely advisory for public buyers. Correction: National strategies under Article 7 are mandatory for Member States and serve as the primary filter for resource allocation. Public-sector buyers must align their projects with these strategies to be eligible for the "endeavour" support under Article 9(3). Ignoring the national strategy significantly reduces the chances of securing compute.

Misconception 3: "Public sector AI" only refers to administrative tools. Correction: The scope is broad. As per Recital 22 and Article 4(7), it includes critical domains like healthcare, energy, and border management, as well as projects that improve decision-making and simplify administrative procedures. It covers the full lifecycle from development to deployment.

Misconception 4: The "endeavour" clause means public projects will be ignored. Correction: Not at all. Article 9(3) explicitly lists public-sector AI alongside industrial and physical AI as a priority area for the Union and Member States. The "endeavour" language reflects the reality of finite resources, not a lack of political will. Public projects remain a core component of the Cloud and AI Leadership Initiatives.

Related

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