Summary Under the proposed Cloud and AI Development Act (CADA), "cloud sovereignty" and "AI sovereignty" are not distinct regulatory silos but integrated layers of a single "Union assurance" framework. Cloud sovereignty (Title IV, Article 16) focuses on jurisdictional control, data localization, and protection against third-country legal extraterritoriality. AI sovereignty (Title II, Articles 3–4) centers on securing the computational capacity, algorithmic independence, and supply chain resilience required to train and run models without foreign dependency. While cloud sovereignty ensures data remains under EU jurisdiction, AI sovereignty ensures the "intelligence" processing that data is built on EU-designed hardware and open, auditable code. The Commission's explanatory memorandum explicitly states that while the AI Act governs AI systems, it "does not cover aspects of sovereignty," a gap CADA is designed to fill by addressing the infrastructure and compute beneath the model.

Detail

The distinction between cloud sovereignty and AI sovereignty is fundamental to the architecture of the proposed CADA (COM(2026) 502 final). While often conflated in public discourse, the proposal treats them as complementary but distinct layers of a broader strategy for "technological autonomy." CADA does not create a separate "AI sovereignty" law; rather, it embeds AI-specific resilience measures within a holistic framework that spans from the physical data centre to the algorithmic model.

Cloud Sovereignty: Control Over Data, Infrastructure, and Legal Jurisdiction

Cloud sovereignty, as defined in CADA's Title IV (specifically Article 16), is primarily about jurisdictional control, data protection, and the prevention of unauthorized third-country access. It addresses the risks identified in the explanatory memorandum regarding dependence on providers subject to laws with extraterritorial effects, which may conflict with EU fundamental rights.

The core mechanism is the Union cloud computing sovereignty framework, which establishes four cumulative "Union assurance levels" (Level 1 to Level 4). These levels are defined in Annex II and escalate in strictness:

  • Level 1 (Baseline): Requires the provider to be established in the Union, with infrastructure and assets located in the Union. Crucially, customer data (including metadata and telemetry) must remain exclusively within the Union unless the public sector body explicitly requires otherwise (Annex II, Section 1.1(c)).
  • Levels 2 & 3 (Enhanced Assurance): These levels introduce stricter requirements for personnel (Union citizenship is conditional at Level 2 if the public body requires it, but mandatory at Level 3), cybersecurity certification (at least "substantial" assurance under a European scheme), and supply chain transparency.
  • Level 4 (High Assurance): The highest tier requires that sensitive data identified via risk assessment remains exclusively in the Union, personnel must be Union citizens (with security clearances where appropriate), and the service must obtain a European cybersecurity certificate of at least assurance level "high" (Annex II, Section 4.1(e)).

Cloud sovereignty is essentially about where the data lives, who controls the physical assets, and which legal jurisdiction applies. It mitigates the risk of foreign laws (such as the US CLOUD Act) compelling a provider to hand over EU-hosted data by ensuring that no third country can legally or technically override EU operational autonomy.

AI Sovereignty: Control Over Compute, Models, and Innovation

AI sovereignty, while less explicitly defined as a standalone term in the enacting articles, is operationalized through the Cloud and AI Leadership Initiatives (Title II, Articles 3–4). Here, the focus shifts from data storage to computational power, algorithmic control, and industrial competitiveness.

The explanatory memorandum highlights that AI unlocks opportunities through "automation and data-driven decision-making," but this requires "computational capabilities" that are currently scarce in the EU. AI sovereignty, therefore, is about:

  1. Compute Capacity: Ensuring sufficient, energy-efficient compute resources are available in the EU to train and run AI models without relying on foreign hyperscalers. This is addressed through the deployment of data centres, "acceleration zones," and the goal to "triple EU capacity in the next five-to-seven years" (Explanatory Memorandum, p. 2).
  2. Model and Algorithm Control: Supporting the development of European frontier AI, physical AI, and industrial AI (Article 4). This includes ensuring that data generated in the EU is not used to train AI systems operated by third countries (Annex II, Section 2.1(f) and 3.1(f)).
  3. Supply Chain Resilience: Reducing dependence on non-EU semiconductor and hardware suppliers for AI-optimized servers and accelerators (Article 4, operational objective 2). The proposal explicitly supports "AI-optimised servers and baseline software based on processors, accelerators and quantum accelerators designed and manufactured in the Union."

In short, while cloud sovereignty protects the data, AI sovereignty protects the processing power and intellectual property behind the algorithms that use that data.

The Intersection: Why CADA Covers Both

CADA recognizes that AI cannot function without cloud infrastructure, and cloud infrastructure is increasingly defined by AI workloads. The proposal explicitly links the two through specific criteria that bridge the gap between storage and computation:

  • Data for AI: The criteria for Union Assurance Levels 2 and 3 explicitly state that data generated by using the audited service must not be used to train or fine-tune any AI system operated by a third country or a legal entity established in a third country (Annex II, Section 2.1(f) and 3.1(f)). This bridges cloud data localization with AI model training sovereignty.
  • Compute for Sovereignty: The proposal aims to "triple EU capacity in the next five-to-seven years" to meet AI demands (Explanatory Memorandum, p. 2). Without this compute sovereignty, EU organizations cannot independently run sovereign AI models, rendering cloud sovereignty insufficient for advanced AI use cases.
  • Sovereign AI Capabilities: The explanatory memorandum notes that the EU must "maintain a foothold in areas where technological sovereignty is required, such as security and encryption ('sovereign cloud' solutions) and thus reduce critical external dependencies by strengthening homegrown cloud and AI capabilities." This indicates that cloud and AI sovereignty are two sides of the same coin: one secures the storage, the other secures the intelligence.

The Commission explicitly frames CADA as reinforcing the AI Act, noting that the AI Act "does not cover aspects of sovereignty." CADA fills this gap by regulating the infrastructure and supply chain beneath the AI layer.

What this means for you

For CTOs, architects, and SMEs, distinguishing between these two concepts is critical for compliance planning and technology selection under the proposed CADA.

1. Procurement Strategy for Public Sector and Critical Infrastructure If you are a public sector body or operate in a sector listed under Annex I of the NIS2 Directive, you must conduct risk assessments (Article 29) to determine which Union Assurance Level is required.

  • Cloud Sovereignty Check: Ensure your provider is recognized under Article 17. For most public services, Level 1 is the minimum. For critical public order activities, you may need Level 2, 3, or 4.
  • AI Sovereignty Check: If your use case involves AI, verify that the provider's infrastructure meets the compute requirements and that your data is not being siphoned off to train third-party models. Look for providers participating in the "Cloud and AI Leadership Initiatives" or those using EU-designed processors (Article 4).

2. Vendor Lock-in and Multi-Cloud Strategies CADA encourages multi-cloud strategies to enhance resilience (Recital 65). When evaluating vendors, do not just ask, "Where is your data center?" (Cloud Sovereignty). Also ask, "Where is your AI compute located, and who owns the model weights?" (AI Sovereignty).

  • Action: Audit your stack. If you are using a third-country cloud provider for storage but an EU provider for AI inference, you may still face sovereignty gaps if the underlying compute or model training relies on non-EU infrastructure.

3. SMEs and Open Source CADA promotes open source as a lever for sovereignty (Articles 41–44). For SMEs, leveraging open-source AI models and cloud stacks can reduce dependency on proprietary, third-country-controlled technologies.

  • Action: Prioritize open-source solutions listed in the EU Open Source Solutions Catalogue (Article 43) where feasible. This enhances both cloud and AI sovereignty by ensuring code transparency and reducing vendor lock-in.

4. Compliance Costs and Audits Be prepared for audits. Cloud sovereignty compliance (Levels 2–4) requires independent third-party audits (Article 20). AI sovereignty compliance may involve demonstrating that your AI workflows are integrated with EU compute resources and that your data governance prevents cross-border training leakage. Note that for SMEs, the EU statement of conformity for Level 1 is directly and automatically recognized in all Member States without prior recognition by the national competent authority (Article 17(3)).

Common misconceptions

Misconception 1: "Cloud sovereignty is just about data localization."

  • Reality: While data localization is a key component (Annex II), cloud sovereignty also encompasses legal jurisdiction, personnel citizenship (mandatory at Levels 3 and 4), cybersecurity certification, and supply chain transparency. A provider can host data in the EU but still fail Level 3 if their ultimate beneficial owners are subject to third-country laws that could compromise operational autonomy.

Misconception 2: "AI sovereignty is only about building European AI models."

  • Reality: AI sovereignty is equally about the infrastructure that runs those models. CADA emphasizes the need for EU-designed processors, accelerators, and quantum technologies (Article 4). Without sovereign compute, even a European AI model is vulnerable if it must be trained or run on foreign hardware.

Misconception 3: "CADA creates two separate compliance tracks."

  • Reality: CADA integrates cloud and AI into a single regulatory framework. The Union Assurance Levels (Article 16) apply to cloud computing services, but the criteria explicitly include AI-related safeguards (e.g., preventing data use for third-country AI training). You do not comply with "cloud rules" and "AI rules" separately; you comply with a unified sovereignty framework that covers both.

Misconception 4: "Sovereignty means banning all third-country providers."

  • Reality: CADA is not a blanket ban. It establishes a risk-based approach. Third-country providers can qualify for Union Assurance Level 3 if the Commission determines their country provides sufficient safeguards against unauthorized access and service disruption (Article 18). However, Level 4 generally requires strict EU control, making it difficult for third-country entities to qualify for the most sensitive use cases.

Misconception 5: "The AI Act already covers sovereignty."

  • Reality: The AI Act harmonises rules for AI systems and general-purpose AI models to be placed on the EU market, ensuring a high level of protection of health, safety and fundamental rights. It "does not cover aspects of sovereignty." CADA is the specific instrument proposed to address third-country control, extraterritorial access, and operational continuity of the underlying infrastructure.

Official sources

Related

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