Sovereign Gemini: AI that doesn't leave the country

The biggest announcement Google made at I/O Connect India is the availability of the Gemini model on Google Distributed Cloud — cloud infrastructure running directly inside Indian data centers. For regulated sectors such as banking, government, or healthcare, this means one crucial thing: all prompts, model outputs, and even the model weights themselves remain under the operator's control and never leave Indian territory.

The term "sovereign AI" refers to a system that is subject to local legislation, data jurisdiction, and security standards of a given country. For Google, this is a strategic move — India, with its 1.4 billion people, represents one of the largest AI markets in the world, and local regulations increasingly require sensitive data to remain on domestic soil.

In addition to the full-fledged Gemini, Gemini 3.5 Flash will also be available to Indian companies and startups through the Gemini Enterprise Agent Platform and the Gemini Enterprise app. Google guarantees that all machine learning will be processed exclusively in Indian data centers.

For Czech readers, this development is interesting as a parallel to the European demand for data sovereignty. The EU AI Act imposes similar requirements for data protection and transparency — with this move, Google is testing a model in India that could be replicable for the European market. Czech companies working with sensitive data would welcome such a solution, especially given concerns about storing data on US servers.

AI goes to school: 56 hours of free coursework and an assistant for teachers

The educational pillar of India's AI strategy is striking in its scope. Google DeepMind has launched a free 56-hour AI Research Foundations course that teaches participants how to build and fine-tune large language models. The course already has over 38,000 enrolled students from around the world. The Indian Institute of Science, IISc Bengaluru, is incorporating it into its curriculum, and the organization NASSCOM will make it available through the FutureSkills Prime platform.

A practical impact for schools comes from the tool ATL Saathi — an assistant powered by the Gemini model that helps teachers in so-called Atal Tinkering Labs, which are laboratories equipped for technical experimentation in Indian schools. In the pilot phase, the tool will be deployed in 100 schools, with ambitions to expand to 10,000 schools.

This is a demonstration of how AI can serve not only the technological elite but also the public education system. In the Czech Republic, such large-scale programs for teachers are still lacking — although the Ministry of Education is working on framework educational programs that include digital competencies, practical tools like ATL Saathi are not yet found in Czech schools.

Healthcare powered by AI: Leprosy and reproductive health

Google placed strong emphasis on healthcare applications. Scientists from the prestigious AIIMS (All India Institute of Medical Sciences) are expanding their collaboration on open MedGemma models — a specialized version of the Gemma model designed for medical use. The goal is to build models focused specifically on Indian health challenges — in particular, leprosy diagnostics and sexual and reproductive health.

The models will process both image and text inputs, and Google plans to make them available to the entire Indian developer community as open-source. India's National Health Authority recently used the Gemma 4 model in conjunction with Google's open-source Medical Data Toolkit to build a new version of the Aarogya Setu 2.0 app — a mobile platform for tracking citizens' health status that served hundreds of millions of Indians during the COVID pandemic.

In the European context, this approach is also relevant for the Czech Republic — Czech hospitals and research institutions could benefit from similar open-source medical models, especially in areas where there is a shortage of specialized doctors. However, a model built on AIIMS data will primarily be trained on the Indian population, so its direct application in Czech healthcare would require additional testing.

Gemini speaks 25 Indian languages — including Sanskrit

India's linguistic diversity is enormous — 22 official languages and hundreds of dialects are spoken there. Google is responding by expanding Gemini Live to more than 25 Indian languages and dialects, including historical languages like Sanskrit, or regional variants such as Bhojpuri and Maithili.

This is a practical demonstration that large language models are no longer the domain of English and a few major world languages. For Czechs, this is an encouraging signal — Gemini and other Google models support Czech at a good level, and the expansion into more languages shows that Google is investing in linguistic diversity systematically.

Part of the language strategy is also the Vaani project, on which Google is collaborating with IISc Bengaluru. The result is open-source speech and image datasets covering 109 Indian languages. These datasets are freely available to researchers worldwide and can also help in developing models for other underrepresented languages.

What can Europe take from this?

In India, Google is testing a model that combines cloud sovereignty, the public sector, education, and healthcare into a single integrated AI ecosystem. For Europe — which, under the EU AI Act, is seeking a balance between regulation and innovation — this is a valuable precedent.

Czech companies and public institutions face similar questions as their Indian counterparts: How to use AI in healthcare without sensitive patient data leaving the country? How to train teachers and students on technology that changes every month? And how to ensure AI understands smaller languages too, not just English?

The Indian example shows that the path lies through open models, local data centers, and massive investment in education — not through waiting for perfect regulation. For Google, India is the second most important market after the US, and investments worth billions of dollars reflect that. It is expected that Google will offer a similar sovereign AI model to other countries, including European states.