Summary Under the proposed Cloud and AI Development Act (CADA), "physical AI" refers to AI systems and models capable of perceiving the physical environment and executing complex actions within it. As proposed, CADA establishes Operational Objective 4 to accelerate the development of a "European physical AI stack," specifically targeting robotics, autonomous vehicles, and drones. This initiative is operationalised through Grand Challenge 4 in Annex I, which focuses on deploying these systems in unstructured environments with minimal human supervision. While CADA drives the strategic development and capacity-building for these technologies, it does not replace the safety and conformity requirements of the EU AI Act, which would still govern the market placement of physical AI systems.

Detail

The Cloud and AI Development Act (CADA), formally the Proposal for a Regulation establishing a framework of measures for strengthening Europe's cloud and AI ecosystem (COM(2026) 502 final), identifies physical AI as a critical frontier for European technological sovereignty. Unlike traditional software-based AI, physical AI integrates digital intelligence into tangible systems that interact with the real world. The proposal frames this not merely as a technological upgrade but as a strategic necessity to reduce external dependencies and foster industrial competitiveness.

Defining Physical AI in the CADA Proposal

While the general definitions in Article 2 of the proposal do not contain a standalone definition for "physical AI," the concept is explicitly delineated in the recitals and operational objectives. Recital 17 provides the authoritative description within the proposal, defining physical AI as "AI systems and models capable of perceiving the physical environment and executing complex actions within that environment."

The text highlights that this technology represents a frontier where "advanced digital intelligence is integrated into tangible systems." The proposal explicitly lists key examples of such systems:

  • Robotics
  • Autonomous drones
  • Self-driving vehicles

The proposal argues that physical AI is "essential to mitigate external dependencies and foster industrial competitiveness and strategic autonomy." Because these systems rely heavily on real-time data processing and low-latency decision-making, the text notes that physical AI "requires a dedicated approach to data and computing infrastructure." Consequently, the proposal calls for "targeted support for the testing and validation of physical AI models and systems in diverse real-world environments to ensure their robustness and reliability."

Operational Objective 4: Building the European Physical AI Stack

The core legislative mechanism for advancing physical AI is found in Article 4, which sets out the operational objectives of the Cloud and AI Leadership Initiatives. Specifically, Article 4(4) mandates that these initiatives shall:

  1. Accelerate the development of a European physical AI stack: This includes supporting "model training and system development and deployment, in particular for robotics and autonomous vehicles and drones." The objective is to create a sovereign ecosystem of software, hardware, and standards that reduces reliance on non-EU technologies.
  2. Facilitate data access: The initiatives must "facilitate access to, and the collection and preparation of, specific datasets for physical AI." High-quality, domain-specific data is a prerequisite for training models that can safely navigate complex physical environments.
  3. Support real-world validation: The initiatives shall "support the development, testing and validation in real-world environments of physical AI models and systems." This moves beyond simulation, ensuring that systems are robust before deployment.

This objective is designed to bridge the gap between theoretical AI models and their practical, safe application. By focusing on a "European physical AI stack," the proposal aims to ensure that the underlying hardware (processors, sensors) and software (models, middleware) are designed and manufactured within the Union, thereby strengthening the supply chain.

Grand Challenge 4: Strategic Deployment in Unstructured Environments

The implementation of Article 4(4) is driven by the "Grand Challenges" outlined in Annex I of the proposal. Grand Challenge 4 is dedicated to "Developing advanced physical AI models and systems that operate autonomously and safely for delivering robust, manipulation and navigation in unstructured environments."

The text specifies that the focus of this challenge will be on "co-designing software and its underlying hardware architectures." It aims to combine "frontier AI techniques with world models supporting physical reasoning." The ultimate goal is to deliver "robust manipulation, navigation, and interaction capabilities with minimal human supervision."

The potential applications explicitly mentioned in Annex I include:

  • Autonomous robots
  • Industrial systems
  • Drones operating in dynamic real-world environments

This challenge recognises that unstructured environments (such as disaster zones, complex urban settings, or dynamic industrial floors) present unique challenges that require AI systems to reason physically and adapt in real-time, rather than following pre-programmed scripts.

Sovereignty, Security, and the AI Act Intersection

The development of physical AI under CADA is intrinsically linked to the broader sovereignty framework. Recital 16 and Recital 20 emphasise that frontier and physical AI technologies are "critical strategic assets." The proposal seeks to ensure the Union has the capacity to develop and govern these technologies in alignment with Union values and safety standards.

Recital 20 further notes that the Union should foster "highly secured computing infrastructures for the training, testing and deployment of defence-related AI models and systems," which often overlap with physical AI capabilities (e.g., autonomous defense systems). This underscores the need for public-sector bodies to consider the sovereignty of the underlying cloud infrastructure used to train and operate these physical systems.

It is crucial to distinguish CADA's role from that of the EU AI Act (Regulation (EU) 2024/1689). CADA is an industrial and strategic policy tool aimed at building and deploying European capabilities. The AI Act is a product-safety and fundamental-rights regulation. Many physical AI systems, such as those used in autonomous vehicles or industrial robotics, will likely fall under the "high-risk" category of the AI Act. Therefore, while CADA encourages the development of these systems through funding and strategic initiatives (like Grand Challenge 4), any physical AI system placed on the EU market must still comply with the strict conformity assessment, risk management, and transparency requirements of the AI Act. CADA provides the capacity and strategy; the AI Act provides the safety rules.

What this means for you

For public-sector procurement officers, digital transformation leaders, and industry stakeholders, the CADA proposal has several practical implications regarding physical AI:

  1. Strategic Sourcing and the European Stack: When procuring advanced robotics, autonomous drones, or self-driving solutions, be aware that the EU is actively fostering a domestic supply chain. CADA's push for a "European physical AI stack" may lead to a growing market of EU-based providers offering sovereign, compliant alternatives to non-EU incumbents. Procurement strategies should anticipate a shift towards these emerging European capabilities.
  2. Sovereignty Assurance Levels (UALs): If your public-sector body uses cloud services to train or operate physical AI models, you must adhere to the Union Assurance Levels established in CADA's Annex II. For high-criticality applications (e.g., defense, emergency response, critical infrastructure), you may be required to procure services at UAL 3 or 4. These levels impose strict requirements on data localization, personnel citizenship (Union citizens), and freedom from third-country control, ensuring that the physical AI system's "brain" is not subject to foreign interference.
  3. Access to Testing Infrastructure: CADA supports the creation of testing and validation environments for physical AI in real-world conditions (Article 4(4)(c)). Public authorities may have access to new national or EU-funded testing facilities, potentially linked to the Experience and Acceleration Centres for AI (Article 5), to validate the safety and reliability of physical AI systems before large-scale deployment.
  4. Dual Compliance: Remember that CADA does not replace the AI Act. Procurement of physical AI systems must still ensure compliance with the AI Act's high-risk requirements, including fundamental rights impact assessments, data governance, and human oversight measures. CADA provides the capacity and strategy; the AI Act provides the safety rules.

Common misconceptions

"CADA regulates the safety of physical AI robots."

  • Reality: CADA is an industrial strategy and capacity-building regulation. It does not set safety standards for the devices themselves. Safety, liability, and market access rules for physical AI systems are governed by the AI Act, the Machinery Regulation, and other sector-specific product safety laws. CADA focuses on the infrastructure and sovereignty of the AI stack.

"Physical AI only refers to manufacturing robots."

  • Reality: As defined in CADA's Grand Challenge 4, physical AI includes a broad range of systems operating in unstructured environments. This explicitly includes autonomous drones, self-driving vehicles, and complex robotic systems for navigation and manipulation, not just fixed industrial arms.

"Procuring a physical AI system automatically makes it 'sovereign'."

  • Reality: Sovereignty under CADA is determined by the Union Assurance Levels (UALs) of the underlying cloud services, data processing, and control structures. A physical robot may be manufactured in the EU, but if its AI model is trained on non-EU cloud infrastructure subject to third-country laws, it may not meet the higher UALs required for critical public-sector use cases. The "European physical AI stack" is the intended solution to this gap.

Official sources

Related

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