Summary Under the proposed Cloud and AI Development Act (CADA), Article 9(3) establishes a distinct support mechanism for Physical AI, AI industrial innovation, and public sector AI. Unlike Frontier AI priority projects, which benefit from a strict "matching" obligation where the Union matches Member State contributions, the Union and Member States shall only "endeavour to provide sufficient computing resource" for Physical AI projects. This "endeavour" basis reflects the diverse nature of industrial and physical AI use casesβ€”ranging from autonomous robots to dronesβ€”without imposing the rigid multi-state consortium criteria required for Frontier AI. This approach aligns with Grand Challenge 4 in Annex I, ensuring that tangible, real-world AI applications receive necessary infrastructure support while maintaining flexibility for industrial deployment.

Detail

The proposed Cloud and AI Development Act (CADA), COM(2026) 502 final, seeks to address the EU's compute capacity deficit by tailoring support mechanisms to specific strategic domains. While the Act creates a high-profile pathway for "Frontier AI," it simultaneously recognizes that the EU's industrial competitiveness relies heavily on Physical AIβ€”systems that perceive and act upon the physical world. The distinction in support mechanisms between these two categories is codified in Article 9, specifically contrasting the mandatory matching rules for Frontier AI with the "endeavour" obligation for Physical AI.

Defining Physical AI: Grand Challenge 4

To understand the scope of Article 9(3), one must first look at the definition of Physical AI within the CADA framework. Annex I, which outlines the "Grand Challenges" for the Cloud and AI Leadership Initiatives, explicitly identifies Grand Challenge 4: Physical AI.

The proposal defines this challenge as:

"Developing advanced physical AI models and systems that operate autonomously and safely for delivering robust, manipulation and navigation in unstructured environments."

The text further specifies that the focus is on "co-designing software and its underlying hardware architectures" to enable:

  • Robust manipulation, navigation, and interaction capabilities with minimal human supervision.
  • Applications in autonomous robots, industrial systems, and drones operating in dynamic real-world environments.

Unlike purely software-based AI, Physical AI requires a dedicated approach to data and computing infrastructure. As noted in the explanatory memorandum, these systems require "targeted support for the testing and validation of physical AI models and systems in diverse real-world environments." This necessitates not just raw compute power, but access to environments where these systems can be trained and validated safely.

Article 9(3): The "Endeavour" Obligation

Article 9 of the proposal, titled "Computing support for AI projects," is the central provision governing resource allocation. While paragraphs 1 and 2 establish a rigorous framework for Frontier AI, paragraph 3 sets a different standard for Physical AI.

Article 9(3) 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 "endeavour" is legally significant in EU legislative drafting. It indicates a commitment to act with diligence and good faith to achieve a result, but it does not create an absolute, enforceable obligation to deliver a specific quantity of resources in every instance. This contrasts sharply with the language used for Frontier AI.

The rationale for this distinction lies in the nature of the projects:

  1. Diversity of Use Cases: Physical AI spans a wide spectrum, from a single SME developing a specialized industrial robot to a large consortium building autonomous drone fleets. A rigid "matching" formula might not suit all these scenarios.
  2. Industrial Innovation: The provision groups Physical AI with "AI industrial innovation," acknowledging that these projects often emerge from existing industrial ecosystems rather than purely academic or research-led consortia.
  3. Flexibility: The "endeavour" clause allows the Union and Member States to prioritize resources based on evolving needs, strategic sectors, and available capacity, rather than being bound by a fixed matching ratio.

Contrast with Frontier AI Priority Projects

The difference between the support for Physical AI and Frontier AI is stark when comparing Article 9(3) with Article 9(1) and Article 9(2).

Frontier AI Priority Projects (Article 9(1) & (2)) Frontier AI projects are defined by Article 8 as "pioneering projects" focused on scaling up frontier AI technologies. To qualify, they must:

  • Be undertaken by a European digital infrastructure consortium or eligible legal entity.
  • Involve the participation of at least three Member States.
  • Pool computing time and resources.

For these projects, Article 9(2) imposes a strict obligation:

"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 creates a matching mechanism: if Member States contribute resources, the Union must match them (subject to availability). This is a high-bar, high-reward model designed to catalyze massive, cross-border research initiatives in foundational models.

Physical AI Projects (Article 9(3)) Physical AI projects do not need to meet the Article 8 criteria (e.g., the three-Member State consortium requirement) to be eligible for support under Article 9(3). Instead, they fall under the broader "endeavour" clause.

  • No Matching Guarantee: There is no statutory requirement for the Union to match Member State contributions for Physical AI.
  • Broader Eligibility: Support is available for "AI industrial innovation" and "public sector AI" alongside Physical AI, meaning a wider range of actors (including single Member State initiatives or private industrial players) can benefit.
  • Resource Allocation: While the Union is not obliged to match, it is committed to endeavouring to provide sufficient resources. This suggests that support will be available but may be allocated through national strategies, specific calls, or the Centres for AI network rather than a centralized matching pool.

Link to Operational Objectives and Leadership Initiatives

The support for Physical AI under Article 9(3) is not an isolated provision; it is the operational engine for Operational Objective 4 of the Cloud and AI Leadership Initiatives (Article 4(4)). This objective mandates the Union to:

"Advance Union's capabilities in physical AI models and systems and foster their deployment across the Union's strategic sectors."

Article 9(3) provides the computational fuel for this objective. By ensuring that "sufficient computing resource" is available, the Act enables the testing and validation phases described in Grand Challenge 4. Without this compute support, the "co-designing of software and hardware" and the "robust manipulation" capabilities central to Physical AI would be stifled by the lack of access to high-performance computing (HPC) infrastructure.

Furthermore, the provision supports the Experience and Acceleration Centres for AI (Article 5), which are tasked with supporting the integration and scaling-up of AI use cases in strategic industrial sectors. These centres will likely serve as the primary interface for Physical AI projects seeking to access the compute resources promised under Article 9(3).

What this means for you

For CTOs, industrial AI developers, and public sector bodies, the "endeavour" clause in Article 9(3) has specific strategic implications.

1. Strategic Planning for Compute Access

Unlike Frontier AI projects, where you can rely on a predictable matching mechanism if you form a multi-state consortium, Physical AI projects must adopt a more flexible strategy.

  • Engage Early with National Strategies: Since the "endeavour" obligation is shared between the Union and Member States, national implementation is key. Review the national cloud and AI strategies (required under Article 7) to see how your Member State plans to allocate compute for Physical AI.
  • Leverage the Centres for AI: The Experience and Acceleration Centres for AI (Article 5) are your primary access point. They are designed to bridge the gap between research and industrial deployment. Engage with these centres to understand how they will distribute the "sufficient computing resources" promised in Article 9(3).

2. Project Design and Eligibility

  • Focus on Industrial Value: When applying for support, emphasize the "industrial innovation" aspect of your project. Article 9(3) explicitly groups Physical AI with industrial innovation. Demonstrating how your project contributes to the EU's industrial competitiveness (e.g., in automotive, manufacturing, or logistics) will strengthen your case for resource allocation.
  • Real-World Validation: Highlight the need for "testing and validation in diverse real-world environments." The Act recognizes that Physical AI requires more than just raw compute; it requires the right environments. Proposals that include access to test beds or pilot lines (as mentioned in Grand Challenge 4) may be prioritized.

3. Managing Expectations

  • No Guaranteed Match: Do not assume that the Union will match your national or private investment. The "endeavour" language means that while the political will exists, the actual allocation depends on available capacity and strategic prioritization.
  • Diverse Pathways: Be prepared to access compute through multiple channels: national HPC centers, the EuroCloud Federation (Article 34), or specific calls for physical AI projects, rather than a single centralized "Frontier AI" pool.

Common misconceptions

"Physical AI projects get the same matched compute as Frontier AI."

  • Reality: This is incorrect. Only Frontier AI priority projects (meeting Article 8 criteria) benefit from the mandatory "matching" obligation in Article 9(2). Physical AI projects are supported under Article 9(3) on an "endeavour" basis, which is a commitment to strive for sufficiency but not a guarantee of a 1:1 match.

"Physical AI is just about robotics."

  • Reality: While robotics is a core component, Grand Challenge 4 explicitly includes autonomous drones, industrial systems, and any system that "perceives the physical environment and executes complex actions." This covers a broad range of tangible AI applications beyond just humanoid or industrial robots.

"You need a three-Member State consortium to get Physical AI support."

  • Reality: The three-Member State requirement applies strictly to Frontier AI priority projects under Article 8. Physical AI projects under Article 9(3) do not have this eligibility barrier, allowing for national or single-entity initiatives to access support.

"The 'endeavour' clause means no support is guaranteed."

  • Reality: While not a strict entitlement, "endeavour" is a binding legal obligation to act with diligence. The Union and Member States must actively work to provide sufficient resources. This is a stronger commitment than a mere "consideration" but weaker than a "shall provide" mandate.

Related

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