This category covers general-purpose AI services: large language models, chat assistants, translation models and generative tools for text, images and speech. Organisations use them to draft and summarise documents, answer questions about internal knowledge, translate content and automate customer support, through a hosted application, an API or a model they run themselves.
The sovereignty question goes beyond where a request is processed. Who trains the model and on what data, whether you can run the model weights on your own infrastructure, and how tightly your prompts, fine-tuning data and embeddings are tied to one provider together determine your dependency. Open-weight models can be hosted in a European data centre, while closed models exist only behind the vendor's API. The EU AI Act also sets transparency and risk-management duties for the systems you deploy, regardless of who built the model.
When choosing a European option, look at the model's documented provenance and licence, whether a self-hosted or private deployment is offered, how interchangeable the API is with other providers, and what the provider states about logging and retention of inputs. Compare quality on your own languages and tasks rather than on generic benchmarks.