A country's effort to build AI with its own models, data and computing power.
Sovereign AI means a country building up the ability to make AI with its own data, computing power and people. "Sovereign" means self-governing, like an independent nation.
Those who back it say a country needs AI that knows its own language and culture, and should keep control of important data. India approved a plan in 2024 including at least 10,000 GPUs, and in 2025 France announced €109 billion of private investment in French AI.
In August 2025 Korea picked five teams for its independent AI foundation model project, worth ₩213.6 billion in total; after staged reviews, three remained in August 2026. With this much public money involved, deciding what counts as "independent" and how to measure results matters too.
Think of it like this
At a school festival, some classes order all their food in, while others buy ingredients and cook it themselves. Cooking lets you suit your class's taste and change the menu, but it takes more time and money. Sovereign AI is a country choosing to do its own cooking.
For example
Releasing the models built by teams in a government programme so that anyone can download and use them is one form sovereign AI policy can take.
Common mix-ups
Mix-upSovereign AI means making everything at home, right down to the chips.
ActuallyJapan's national AI computer ABCI 3.0, announced in 2024, was built with 6,128 H200 chips from the American company NVIDIA. India, too, chose to have the state buy GPUs and share them out. Where to draw the line on what must be your own differs from country to country.
Mix-upA high performance score is enough to count as "independent AI".
ActuallyIn Korea's first-round review in January 2026, Naver Cloud placed in the top four of five teams on overall score but was eliminated, judged not to meet the independence standard because it reused an outside open model's weights unchanged in parts that process audio and other inputs (encoders). The government said the problem was not using open models as such, but that the team hadn't shown it had trained the weights itself.
Try it in class
For teachers at the start of a lesson — or for students working on their own.
- Question
- "Our country needs its own AI" versus "we can just use good foreign AI": what does each side care about most?
- Activity
- Split into two groups, one taking "language, culture and control of data" and the other "cost and performance", and make evidence cards. Pull numbers and sentences from the sovereign AI statement (E0431), the selection of the elite teams (E0657) and the first-round review (E0825), then write one line both sides could accept.
Records about this
Source-checked moments from AI history.
All four teams' models were judged to meet the independence criteria; with a newly introduced public user evaluation included, the field narrowed to three
The first eliminations in Korea's largest government-funded AI program
A state-led program to foster open-weight models
The turning point at which sovereign AI in Korea moved from a corporate marketing term to a national policy term. This person became head of the government's AI policy 9 months later