Summary The proposed Cloud and AI Development Act (CADA) and the EU AI Act govern distinct layers of public procurement. The AI Act (Regulation (EU) 2024/1689) regulates the safety and fundamental rights of the AI system itself, imposing obligations on providers and deployers. CADA, as proposed in COM(2026) 502 final, regulates the sovereignty and resilience of the cloud infrastructure and supply chain beneath it. Under CADA, public bodies must conduct risk assessments (Article 29) to determine mandatory Union assurance levels (Article 30) for their cloud purchases and apply Union added-value criteria (Article 32) to support the EU ecosystem. These regimes apply in parallel: a public procurement for a high-risk AI system must simultaneously satisfy the AI Act's safety requirements and CADA's sovereignty requirements.

Detail

The intersection of the Cloud and AI Development Act (CADA) and the Artificial Intelligence Act (AI Act) creates a dual-compliance landscape for public sector procurement. While the AI Act focuses on the "what" (the AI system's risk profile), CADA focuses on the "how" and "from whom" (the strategic autonomy of the underlying infrastructure).

Distinct Regulatory Scopes

The AI Act: System Safety and Risk The AI Act is a product-safety and fundamental-rights regulation. It establishes a risk-based framework:

  • Prohibited Practices: Certain AI uses (e.g., social scoring) are banned outright.
  • High-Risk Systems: AI used in critical sectors (e.g., law enforcement, healthcare, education) must meet strict obligations regarding risk management, data governance, transparency, and human oversight.
  • Deployer Obligations: Public authorities using high-risk AI must conduct fundamental rights impact assessments and ensure human oversight.

Crucially, the AI Act does not regulate the location of data centres, the nationality of the cloud provider, or the supply chain of the underlying hardware. As the CADA explanatory memorandum notes, the AI Act "does not cover aspects of sovereignty."

CADA: Sovereignty and Supply Chain Resilience CADA addresses the gap left by the AI Act by establishing a "Union cloud computing sovereignty framework" (Article 16). Its procurement rules (Title IV, Chapter II, Articles 30–33) are designed to reduce dependence on third-country providers and ensure operational autonomy.

  • Sovereignty Assurance Levels: CADA defines four levels of assurance (1 to 4) based on criteria such as establishment in the EU, location of infrastructure, personnel citizenship, and absence of third-country control (Annex II).
  • Mandatory Procurement Rules: Unlike the AI Act, which applies to the use of AI, CADA imposes specific procurement mandates on contracting authorities.

Key CADA Procurement Obligations

1. Mandatory Assurance Levels (Article 30) CADA Article 30 establishes a tiered procurement requirement based on a prior risk assessment.

  • Baseline Requirement (Article 30(2)): Union entities and public sector bodies whose activities have not been identified as contributing to the preservation of public order must procure cloud computing services recognised as offering at least Union assurance level 1.
  • Public Order Relevance (Article 30(3)): If a risk assessment (conducted under Article 29) determines that an activity contributes to the preservation of public order (e.g., national security, defence, justice, law enforcement, or sectors under the NIS2 Directive), the contracting authority must only procure services recognised as offering Union assurance levels 2, 3, or 4.

2. Union Added Value Criteria (Article 32) CADA Article 32 introduces "Union added value" as a mandatory non-price award criterion for public procurement procedures involving innovative cloud computing services and AI systems.

  • Scope: Contracting authorities must evaluate tenders based on the extent to which the tenderer contributes to strengthening the digital supply chain in the Union, integrates Union technologies, or uses hardware designed/manufactured in the Union.
  • Weighting: These criteria must be "ancillary and not decisive" in the award of the contract (Article 32(2)(d)), ensuring that technical and financial criteria remain primary. However, they must be expressly set out in procurement documents.

3. Innovation and SME Targets (Article 33) Article 33 requires Member States to monitor procurement of innovation in cloud and AI. It sets an objective that at least 25% of such procurement be awarded to innovative SMEs. Member States must include plans to achieve this objective in their national cloud and AI strategies (Article 7).

Parallel Application: How They Work Together

In practice, a single procurement process often triggers both regimes. Consider a public hospital procuring an AI system for diagnostic support:

  1. AI Act Compliance (The System): The hospital must ensure the AI system is classified correctly. If it is "high-risk," the provider must provide a conformity assessment, technical documentation, and a CE marking. The hospital must conduct a fundamental rights impact assessment and ensure human oversight.
  2. CADA Compliance (The Infrastructure): The hospital must first conduct a risk assessment under Article 29 to determine if the diagnostic activity contributes to "public order."
    • If yes, the hospital must procure the cloud hosting for this AI system from a provider recognised at Level 2, 3, or 4.
    • If no, the hospital must procure at least Level 1.
    • Additionally, if the procurement is for an innovative solution, the hospital must apply the Union added-value criteria (Article 32) to the tender evaluation.

The AI Act ensures the tool is safe; CADA ensures the foundation (cloud) is sovereign and the supply chain is resilient.

What this means for you

For in-house counsel, procurement officers, and compliance teams, the intersection of CADA and the AI Act requires a dual-track strategy.

1. Update Procurement Templates and Evaluation Matrices

  • Sovereignty Clauses: Contracts must explicitly require the provider to maintain their recognised Union assurance level. Include termination clauses for material changes that affect recognition (CADA Article 23).
  • Added-Value Scoring: Revise tender evaluation matrices to include Article 32 criteria. Ensure these are weighted as non-price criteria and linked to the subject matter of the contract.
  • Assurance Level Verification: Require bidders to provide proof of recognition (e.g., entry in the central repository under Article 22) before contract award.

2. Conduct Dual Risk Assessments

  • AI Act Risk: Classify the AI system. Is it high-risk? If so, prepare for the provider's conformity assessment documentation and your own deployer obligations (human oversight, logging).
  • CADA Risk: Conduct the CADA-mandated risk assessment (Article 29) to determine if your use case impacts "public order." This assessment dictates the minimum assurance level you can procure. Note that this assessment must be carried out within one year of CADA's entry into force and repeated every two years (Article 29(1)).

3. Monitor SME and Innovation Targets Track procurement spend on innovative cloud and AI solutions. CADA Article 33(4) sets an aspirational target of 25% awards to innovative SMEs. While this is an objective rather than a strict quota, Member States must report on progress. Ensure your lotting strategies facilitate SME participation to meet national strategy goals.

4. Prepare for Penalties and Enforcement

  • AI Act: Penalties for non-compliance with high-risk AI obligations can reach up to €35 million or 7% of total worldwide annual turnover for prohibited practices, and up to €15 million or 3% for other breaches (AI Act Article 99).
  • CADA: CADA Article 24 requires Member States to lay down rules on penalties for infringements of the sovereignty framework. These must be "effective, proportionate and dissuasive." Failure to procure at the required assurance level could expose the public body to regulatory censure and legal challenges.

Common misconceptions

Misconception 1: CADA replaces the AI Act for AI procurement. Reality: CADA does not regulate the safety or ethical operation of the AI model itself. It regulates the cloud infrastructure and the strategic sourcing of the service. You still need full AI Act compliance for the AI system, regardless of which cloud provider hosts it.

Misconception 2: Union Added Value (Article 32) is a quota for EU-made hardware. Reality: Article 32 criteria are non-price award criteria, not quotas. They must be "ancillary and not decisive" (Article 32(2)(d)). You cannot reject a technically superior bid solely because it uses non-EU hardware, but you can award more points to bids that demonstrate stronger EU supply chain integration.

Misconception 3: Only defense and intelligence need high assurance levels. Reality: CADA Article 29 requires risk assessments for any activity contributing to the preservation of public order. This includes sectors listed in Annex I or II of the NIS2 Directive, as well as justice, internal security, and external border management. Healthcare, energy, and transport may also trigger higher assurance requirements depending on the specific risk assessment outcome.

Misconception 4: The AI Act and CADA have the same entry-into-force dates. Reality: The AI Act is already adopted and has phased application dates (prohibitions from Feb 2025, general application from Aug 2026). CADA is a proposal. Its provisions, including the procurement rules in Articles 30–33, are not yet in force and may change during the legislative procedure.

Official sources

Related

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