For years, India was the market the world’s AI companies wanted to sell to. In 2026, it is becoming a place where AI is built. A state-backed push to create home-grown foundation models , big infrastructure investments from Indian and global players, and a much louder voice in shaping global AI rules have changed the conversation. The question is no longer whether India will use AI. It is how much of the AI the world uses will be made in India.



The IndiaAI Mission and the push for Sovereign Models




The IndiaAI Mission was launched in March 2024 by the Government of India and is aimed at building an ecosystem of AI in India that is backed by a five-year outlay of Rs 10,371 crore. The Mission will subsidise access to a growing national pool of GPUs and will also support a number of selected startups and research consortia in the development of foundational capabilities.




The most visible outcome to date was in November 2026 when the Mission presented three indigenous AI models in New Delhi: EKA-1 from Soket Labs, Bengaluru, and Dhīr 2B and Dhīr 5B from ZenteiQ AiTech Innovations. Both companies were selected as part of the IndiaAI Innovation Centre (Foundation Models) initiative which was aimed at developing indigenous large language models, as per the Press Information Bureau (PIB).




The models are for different things. EKA-1 is a foundation model of around 26B parameters. As per the launch coverage, it's aimed at maths, coding, reasoning and agentic work. There are reports it supports all 22 of the scheduled Indian languages which is significant in a country where linguistic diversity has for so long been a barrier to the reach of English-only systems.




The Dhīr models are smaller 2B and 5B parameters. ZenteiQ describes them as open source, supporting English and 10 Indian languages and can be deployed as secure on-premise multilingual assistants. They were reportedly trained from scratch as a mixture of expert models on Google Cloud TPUs. The key feature they offer is the ability to run directly on a mobile phone without requiring an internet connection. That could be a game changer for bringing AI to people in areas with poor connectivity and also speaks to the need of government bodies and regulated enterprises to have their data remain within their own environment. The ministry has specifically cited use in defence, government, industry and education.




Indigenous models reduce reliance on foreign tech in key areas, and the open-source release enables Indian developers, researchers and public institutions to use the models without any licensing restrictions. The small size and offline capability of the models is also very appropriate to India's situation given that so much of the population is on low-end devices and has very poor connectivity.




The launch is a significant milestone, but it has to be looked at with some caution. A 26-billion-parameter model and two smaller language models aren't really the sort of innovations that are going to be direct competitors to the huge frontier systems being built by the world's major labs with vastly greater compute. Their value is in different areas like linguistic coverage, data sovereignty, cost-effective deployment and being well suited to particular public sector use cases. It will be independent benchmarking, real-world uptake and continued funding that will decide whether they end up being the bedrock of a sustainable ecosystem.



Infrastructure: Data centres and computing power at scale




Models need to compute and 2026 has seen a flurry of infrastructure announcements. The most significant of these is the proposed AI data centre campus in Telangana with a potential capacity of one gigawatt and an investment of up to $7.4 billion. Cloud and GPU providers such as Vultr are also ramping up capacity in places like Chennai. If these projects go ahead, they would see the balance of India's digital infrastructure shift from general hosting to being specific for training and inference.




However, energy is the key constraint as GW-scale campuses are going to be drawing heavily on the power grid and the water grid. Their success is going to be contingent on having a very reliable supply of renewables and viable siting. That was a major theme of the policy discussions in New Delhi this year.



Global companies and domestic enterprises




Global tech firms have also noticed the scale of India. OpenAI is reportedly in the process of opening new offices in Mumbai and Bengaluru following a huge spike in ChatGPT usage by Indian consumers. India is now one of the largest markets for generative AI tools, and a local presence would appear to be about more than just subscriptions.




Domestic firms are also moving from experimentation to deployment. TCS has reported that its annualised AI revenue is now over $2.3 billion. They're moving to agentic, AI-orchestrated workflows where software agents are carrying out multi-step business processes rather than the traditional human interface. This is very significant for the Indian IT services sector, as automation changes the way code is written, tested and maintained; the industry is repositioning from a labour-based model to one of AI-enabled outcomes. How this will impact employment across the sector is one of the most important open questions in the Indian economy.



Policy and diplomacy: The India AI Impact Summit




India hosted the AI Impact Summit 2026 in New Delhi, which brought together governments, industry and researchers. The Ministry of Electronics and Information Technology described the event as being structured around the ‘Seven Chakras' framework with a focus on AI that is inclusive, trustworthy and energy efficient, and with a very clear focus on the needs of the Global South.




This push comes at a time when a lot of the earlier international discussion on AI governance was driven by the US, Europe and China. India is positioning itself as the convenor for all the countries that are AI adopters rather than AI developers and it's focused on affordability, linguistic inclusion and development outcomes rather than on security per se.




The World Development Report 2026, The Promise of Artificial Intelligence , was launched jointly by the Ministry of Electronics and Information Technology and the World Bank. As per government statements, the report focuses on localisation, domain-specific regulation and integration into the public sector, with agriculture and health as the key areas of application. Both of these are areas where India has a real strength in terms of scale (the size of the population), the nature of the service delivery and the existence of a well-established digital public infrastructure upon which AI can be built.



Challenges ahead for India




There remain several issues. Access to computing power is still limited despite the fact that it's subsidised. There's a lack of training data in many of the Indian languages, which in turn impacts the quality of the model for those languages. It's very hard to retain talent when the global players are offering much attractive packages. The regulatory issues around data protection, accountability, and the use of AI in public services are still in their infancy. And there’s a gap between the announcements and actual deployment. Gigawatt campuses and national models only really matter if they're actually delivered and used at scale.




India's 2026 AI strategy is very much in motion with a clear and coherent plan. There's the IndiaAI Mission, which is backing models that are designed for the Indian context and Indian languages. There's a big investment in the underlying infrastructure to support larger workloads. In the commercial space, both Indian enterprises and global companies are embedding AI more and more. The summit and the collaboration with the World Bank is about setting the global agenda on inclusive AI. The real test in the years ahead will be whether this momentum translates into real capability, in terms of uptake by citizens and businesses, in the quality of the Indian language models and in the robustness of the underlying infrastructure.