Does CADA Article 9 support only frontier AI, or also industrial and physical AI?

Summary No, Article 9 of the proposed Cloud and AI Development Act (CADA) does not limit support exclusively to frontier AI projects. While it imposes a strict, binding obligation on the Union and Member States to match compute resources for designated frontier AI priority projects, it also explicitly requires them to endeavour to provide sufficient computing resources for AI industrial innovation, physical AI, and public sector AI projects. The distinction lies in the legal force: "shall ensure" for frontier AI versus "shall endeavour" for the others, reflecting a tiered strategy for allocating scarce high-performance computing (HPC) capacity.

Detail

The Cloud and AI Development Act (CADA), as proposed in COM(2026) 502 final, addresses the critical bottleneck of compute capacity in the European Union. The proposal recognises that while frontier AI models require massive, coordinated, and guaranteed access to high-performance computing (HPC) to reach state-of-the-art capabilities, other strategic AI domains—specifically industrial applications, physical AI (robotics and autonomous systems), and public sector use cases—are equally vital for the Union's technological sovereignty and economic competitiveness.

Article 9, titled "Computing support for AI projects," delineates these responsibilities through a nuanced, tiered approach. It distinguishes between strict, mandatory commitments for frontier AI and aspirational, best-effort commitments for other strategic AI sectors. This structure ensures that the most resource-intensive, cross-border research projects receive guaranteed support while maintaining flexibility to support a broader ecosystem of innovation.

Binding Obligations for Frontier AI (Article 9(1) and 9(2))

For frontier AI, CADA imposes clear, mandatory duties that create a legally enforceable expectation of resource allocation.

Article 9(1) 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."

This obligation is reinforced by Article 9(2), which mandates a 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."

The use of the term "shall" in these paragraphs creates a binding legal requirement. If a project is recognised as a "frontier AI priority project" under Article 8—which requires broad participation from at least three Member States and focuses on scaling up frontier technologies—the Union and Member States are obligated to allocate compute time. The Union's matching mechanism ensures that national investments in these strategic assets are amplified by EU-level resources, specifically leveraging the European High Performance Computing (EuroHPC) infrastructure. This creates a "grand challenge" environment where the EU commits to co-financing the compute costs of its most ambitious AI research.

Endeavour-Based Support for Industrial, Physical, and Public Sector AI (Article 9(3))

In contrast, Article 9(3) addresses AI industrial innovation, physical AI, and public sector AI projects with softer, non-binding language. The provision 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 key distinction here is the word "endeavour." In EU legislative drafting, this indicates a best-effort obligation rather than a strict guarantee. It signals that while these sectors are prioritised and supported, the allocation of compute resources is not automatic or guaranteed in the same way it is for frontier AI priority projects. Instead, it requires the Union and Member States to actively seek ways to provide access, likely through existing funding programmes, national strategies, and the broader Cloud and AI Leadership Initiatives, but without the rigid "matching" mechanism applied to frontier AI.

This tiered approach reflects the reality of compute scarcity and the differing nature of the workloads. Frontier AI projects often require massive, dedicated clusters that can only be coordinated at the EU level to achieve global competitiveness. Industrial, physical, and public sector AI projects, while critical for the real-world economy and public services, may have more diverse compute needs that can be met through a combination of national resources, commercial cloud providers, and targeted EU support, rather than a single, guaranteed HPC allocation.

Connection to Grand Challenges 4 and 5

The support for industrial, physical, and public sector AI under Article 9(3) is directly linked to the operational objectives of the Cloud and AI Leadership Initiatives, particularly Grand Challenges 4 and 5 outlined in Annex I of the CADA proposal.

Physical AI (Grand Challenge 4): This challenge focuses on developing advanced physical AI models and systems that operate autonomously and safely. These systems, such as autonomous robots, drones, and industrial systems, require robust manipulation and navigation capabilities in unstructured environments. Article 9(3) explicitly names "physical AI" as a beneficiary of compute endeavours, ensuring that the development of these tangible, real-world AI applications receives attention alongside purely digital frontier models. The "endeavour" clause acknowledges that while these projects need significant compute for testing and validation, their resource profiles may differ from the massive training runs of frontier models.
Industrial AI (Grand Challenge 5): This challenge aims to accelerate the development and deployment of European industrial AI across strategic sectors like automotive, manufacturing, healthcare, and energy. These systems often require specialised computing resources and testing facilities to validate AI systems in real-world environments before large-scale deployment. By including "AI industrial innovation" in Article 9(3), CADA ensures that the industrial base is not left behind in the race for compute capacity. The provision supports the creation of sector-specific models that are adaptable to industrial use cases, as described in the Grand Challenge text.
Public Sector AI: While not a separate Grand Challenge number, public sector AI is a core operational objective (Operational Objective 7 under Article 4). It focuses on accelerating the technological development and uptake of AI in critical public domains, such as healthcare and public administration. Article 9(3) ensures that public sector bodies, which may lack the budget to compete with private hyperscalers for compute time, have a designated pathway to seek compute support through Union and Member State endeavours. This aligns with the broader goal of fostering public-sector adoption of cloud and AI services.

What this means for you

For public-sector bodies, industrial innovators, and research consortia, understanding the distinction in Article 9 is crucial for strategic planning, partnership development, and resource allocation.

1. Manage Expectations on Resource Allocation

If your organisation is developing a frontier AI model that meets the strict criteria of Article 8 (e.g., involving multiple Member States, focusing on state-of-the-art scaling, and addressing a "grand challenge"), you can expect a more guaranteed allocation of EuroHPC resources. The Union is legally bound to match national contributions. However, for industrial, physical, or standard public sector AI projects, you should not assume automatic access to EU compute pools. Instead, you must actively engage with national authorities and EU funding programmes to secure resources under the "endeavour" clause.

2. Leverage National Strategies

Member States are required to adopt national cloud and AI strategies (Article 7) that include measures to support the deployment of data centre capacity and high-intensity computing infrastructure. Your organisation should align its AI compute needs with these national strategies to increase the likelihood of receiving support under the "endeavour" clause of Article 9(3). National strategies are the primary vehicle through which Member States will operationalise their best-effort commitments.

3. Collaborate for Scale and Eligibility

To move a project from an "endeavour" to a more binding support structure, consider collaborating with other entities across Member States. If your industrial or physical AI project can be framed as a cross-border initiative that addresses a strategic Union challenge and meets the technical thresholds of Article 8, it may qualify for stronger support mechanisms or even be considered for frontier AI status. The "grand challenge" framework is designed to be flexible enough to encompass transformative industrial projects if they meet the criteria.

4. Utilise Experience and Acceleration Centres

The CADA proposal establishes a network of Experience and Acceleration Centres for AI (Article 5). These centres are tasked with helping organisations access AI technologies and computing resources. Engaging with these local hubs can provide practical pathways to access compute capacity for industrial and public sector AI projects, even if direct EU HPC allocation is not guaranteed. These centres act as the operational interface for the "endeavour" obligations.

Common misconceptions

"Article 9 guarantees compute for all AI projects."

Reality: Only frontier AI priority projects (defined in Article 8) receive a binding guarantee of matched compute resources from the Union (Article 9(2)). For industrial, physical, and public sector AI, the Union and Member States only "endeavour" to provide resources (Article 9(3)), which is a best-effort commitment, not a guarantee.

"Physical AI and Industrial AI are ignored by CADA."

Reality: CADA explicitly names physical AI and AI industrial innovation in Article 9(3) as priorities for compute support. Furthermore, Grand Challenges 4 and 5 in Annex I are dedicated entirely to these sectors, indicating a strong policy commitment to their development. The "endeavour" language reflects the need for flexibility in resource allocation, not a lack of priority.

"Public sector bodies cannot access frontier AI compute."

Reality: While frontier AI compute is primarily targeted at large-scale, cross-border research projects, public sector bodies can participate in these projects if they meet the Article 8 criteria. Additionally, the "endeavour" clause in Article 9(3) specifically includes public sector AI, ensuring that public administrations have a designated pathway to seek compute support for their specific needs, such as healthcare or crisis management applications.

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