Tech, ExplainedData platforms and AI
Concept

Sovereign AI

A country's own models, in its own languages.

HOW IT WORKSA government fundsGPUs in data centresinside the countryText and speechare gathered inthe country'sown languagesA company thecountry choosestrains a model thereGovernment servicesand local companiesuse the modelThe data, the modeland the computersstay under thecountry's lawsA country that controls its own models does not dependon foreign companies' prices, rules or languages.HOW IT WORKSA governmentfunds GPUs indata centresinsidethe countryText andspeech aregathered inthe country'sown languagesA company thecountrychoosestrains amodel thereGovernmentservices andlocalcompanies usethe modelThe data, themodel and thecomputersstay underthe country'slawsA country that controls its own models does not dependon foreign companies' prices, rules or languages.HOW IT WORKSA government funds GPUs in data centresinside the country↓Text and speech are gathered in thecountry's own languages↓A company the country chooses trains amodel there↓Government services and local companiesuse the model↓The data, the model and the computersstay under the country's lawsA country that controls its own models does not depend onforeign companies' prices, rules or languages.
A country funds GPUs in its own data centres, gathers text and speech in its own languages, has a chosen company train a model there, and its services use the model while the data stays at home.
opens a page inside this one a page in the same scene a page in another scene a concept page a label only, with no page

Sovereign AI means AI models, and the computers that train and run them, that are built inside one country and controlled by it. The models are trained on that country's own languages and data. The goal is that the country's government, companies and citizens need not rely on models owned and run by foreign companies. In India, the government's national AI programme has backed this since 2025, by choosing Indian companies to build a sovereign foundation model, a very large model that other products can be built on.

Here is how it works in general. First, the government pays for, or subsidises, thousands of GPUs, the chips that train and run AI models, in data centres inside the country. Companies, universities and start-ups can rent time on these chips for less than the usual price. This is a sovereign AI cloud. Second, text and speech are collected in the country's languages: books, government documents, news and recordings of people speaking. In India, the government's Bhashini platform gathers and shares such data and models for Indian languages. Third, a company that the country chooses trains a model on this data, in those data centres. Finally, government services and local companies use the model. The data they send to it, the model itself and the computers it runs on all stay under the country's laws.

Why do countries want this? The first reason is language. Most widely used models were trained mainly on English text. They are often weaker in languages such as Odia or Assamese, and they may not know local facts and customs. Many of them also split text in Indian scripts into more small pieces than English text, which makes each answer slower and costlier. The second reason is data. Some data must stay inside the country by rule, and laws such as India's DPDP Act set rules on how personal data may be used. The third reason is control. A foreign company could raise its prices, change its rules or stop selling in the country. Sovereign AI has costs too. It needs a great deal of public money, and a country's own model may still trail the best models in the world.

Imagine a farmer who calls a government helpline and asks about crop insurance in Odia. A voice model trained on Odia speech understands her question and answers in Odia.

Consider a state health department that wants a large language model to summarise lakhs of hospital records. It runs the model on a sovereign AI cloud, so the records never leave the country.

Say a country other than India wants the same. An AI group in Abu Dhabi, in the United Arab Emirates, builds sovereign AI clouds and models, each built for and controlled by one country.

How is sovereign AI different from data localisation? Data localisation only requires that data is stored inside the country. A foreign cloud company can meet that rule by running a data centre in India. Sovereign AI goes further: the model itself, and the right to decide how it is trained and used, also belong to the country. It extends the idea behind India's digital public infrastructure, shared national systems such as Aadhaar and UPI that were built and are controlled in India.

Appears in

Data platforms and AI
Tech, Explained is a part of ingrid.fyi. Every page explains one piece of technology in plain words.