Sovereign AI, term by term.

A principle that admits only as many requests for concurrent processing as the system can safely handle — the rest wait in an orderly queue instead of crashing.

The actual, enforceable control over who can access data — not just where it is stored.

The EU’s AI regulation that classifies AI systems by risk and sets concrete duties for high-risk applications.

The process in which a trained AI model responds to a concrete request — the actual ‘use’ of a model.

The time between request and response — for AI services a decisive measure of user experience.

The EU’s cybersecurity directive that extends security duties to many industries — with liability for management.

AI models that run on your own hardware in your own data centre — instead of as a rented cloud API.

An interface that understands the same calls as the OpenAI API — existing applications keep working without a rebuild.

The operating layer in front of the inference engine: it steers access, queue and orchestration and turns a bare engine into a resilient service.

AI whose data, models and operation stay fully under your own control — without dependence on an external provider.

The volume of requests a system successfully processes in a given time — the measure of scalability under load.

The smallest billing and processing unit of a language model — roughly a word fragment. Cloud AI is paid per token.

A US law that can compel US providers to hand over data — even when it is stored in Europe.

An open-source high-performance engine that runs large language models efficiently — the motor, not the operating layer.