The EU AI Act explained for businesses using cloud services
-
About
- Type
- Blog
New - NVIDIA B200 & B300 Dedicated GPU Clusters | Explore GPU clusters
About
Table of contents
Posted on 18.11.2026
Adding an AI feature to a SaaS product, connecting an application to a model API, or running a model on cloud infrastructure can bring responsibilities under the EU Artificial Intelligence Act. What those responsibilities involve depends on the system’s purpose and your role in developing or using it.
The Act sets rules to protect health, safety, and fundamental rights. A system used to assess job applicants, therefore, faces different requirements from one that filters spam. For businesses building and operating cloud applications, the starting point is understanding what the AI does, who is responsible for it, and which rules apply.
The Act distinguishes between businesses that provide AI systems and those that use them.
A provider develops an AI system, or has it developed, and supplies it or puts it into service under its own name. A deployer uses an AI system in a professional context.
A business using a third-party recruitment tool will generally be a deployer. A company building and supplying its own AI recruitment application may be the provider, even if the application uses another company’s model. Using an existing model does not automatically make the application developer only a deployer.
The cloud infrastructure provider has a separate role. Supplying the compute on which an application runs does not, by itself, make that company the provider of the AI system. Responsibilities depend on what each party develops, supplies, and operates, as set out in the Act’s definitions.
The European Commission describes four broad levels of AI risk:
The intended use matters. The same underlying model could support a writing assistant or an application that assesses job candidates, with different regulatory consequences. Businesses need to assess the resulting system and its purpose, rather than relying solely on how the model supplier describes its technology.
Transparency. Providers of systems that interact directly with people must generally ensure those people know they are interacting with AI, unless this is obvious. Providers of systems generating synthetic content also face requirements for machine-readable marking. Deployers have disclosure duties for certain content, including deepfakes. The specific obligations and exceptions depend on the system, the content, and the business’s role.
AI literacy. Providers and deployers must take measures to support AI literacy among staff and others who operate or use systems on their behalf. Those measures should reflect their knowledge, the work involved, and the people affected. In practice, this includes helping staff understand a system’s limitations and when human oversight is needed.
High-risk systems. Providers face more extensive requirements covering risk management, documentation, human oversight, and cybersecurity. Businesses using these systems have their own duties, including following instructions, assigning competent oversight, and monitoring operations. Buying a tool does not transfer every responsibility to its supplier.
Providers of general-purpose AI models face separate obligations. Businesses building applications on those models must still assess their responsibilities for the resulting system.
The Act’s requirements apply in stages. AI literacy provisions and the original prohibitions began applying in February 2025, followed by rules for general-purpose AI models in August 2025. Transparency requirements began applying on 2 August 2026, subject to specific transitional arrangements.
Under the current implementation timetable:
These later deadlines do not postpone obligations already in force.
Start with an inventory of the AI systems your business uses or supplies, including AI features within existing software. Record each system’s purpose, establish whether you act as a provider or deployer, and identify the requirements and deadlines that apply.
For cloud applications, also map the services involved in processing data. Check the model endpoint, connected APIs, and any services retaining prompts or outputs. This helps establish which suppliers and internal teams need to be involved in the assessment.
The AI Act sits alongside existing law. If an AI application processes personal data, GDPR still applies, including requirements around lawful processing, security, and international transfers. Our explanation of GDPR in cloud computing sets out the distinct responsibilities of customers and cloud providers.
An application can store its database in Europe while sending information to an external AI service for processing. Businesses, therefore, need to understand where those requests go, who can access them, and whether the information is retained.
Data sovereignty also concerns which organizations and jurisdictions can exercise control over that data. Location is one consideration; provider ownership, corporate structure, and access arrangements also matter.
The AI Act does not generally require AI to be hosted in Europe. European hosting requirements should be assessed against the laws, contracts and organizational policies relevant to the workload.
Running a suitable model on your own cloud infrastructure gives you more control over the processing environment, model versions, and access permissions. It also means taking responsibility for operating and securing the model and application.
UpCloud GPU Servers support AI training and inference without the need to buy and maintain physical GPU hardware. UpCloud is European-owned and headquartered in Finland, with a corporate structure designed to protect European data.
Infrastructure can support your security and data governance requirements. Meeting AI Act obligations also depends on how the system is designed, documented and used, including the oversight it receives in day-to-day operation.
Explore GPU Servers and European data sovereignty at UpCloud.