Sovereign AI
How it works
Sovereign AI is delivered by combining several pillars under national control: (1) building or securing domestic compute — data centers and "AI factories" with accelerator clusters; (2) collecting and governing local datasets in line with national law; (3) training or fine-tuning foundation models for native languages and context (e.g. SEA-LION, Apertus, Amalia); (4) developing talent and an ecosystem (researchers, startups, universities); (5) policies, public procurement and regulation favoring local solutions. In practice, implementations sit on a spectrum from full autarky to adapting open models and pooling resources regionally, often still relying on foreign hardware and cloud.
Problem solved
It addresses the risk of strategic dependence on a handful of foreign AI providers: lack of control over where national data is processed, which jurisdictions govern the models, and whether access to compute and models could be cut off for geopolitical reasons. It also tackles the under-representation of local languages and cultural context in global models.
Components
Locally owned and operated data centers and accelerator clusters where models are trained and served without offshoring compute.
National datasets collected and governed under local law, serving as the basis for training and fine-tuning models.
Foundation models trained or fine-tuned for local languages and cultural context (e.g. SEA-LION, Apertus, Amalia).
Domestic researchers, startups, universities and training programs building local AI capabilities.
Regulation, public procurement and investment strategies that favor local solutions and data/model sovereignty.
Evolution
NVIDIA’s CEO begins championing sovereign AI, arguing that countries should build systems reflecting their language, knowledge and culture — a message that resonated strongly with European leaders.
National models and projects for local languages emerge (e.g. SEA-LION in Singapore, Apertus in Switzerland, Amalia in Portugal), alongside investment in domestic compute centers.
The EU announces a plan to invest roughly €20 billion in three to five AI gigafactories, plus the Digital Europe Programme, backing regional compute sovereignty.
More countries launch dedicated funds and centers (e.g. the UK’s £500 million sovereign-AI fund), cementing sovereign AI as part of industrial policy.