Why India Needs to Build Its Own Foundational AI Model
Bernstein says India risks dependence on US and Chinese AI systems unless it builds its own foundational model, sharpening the debate over sovereign AI, compute access and who controls the country’s next technology layer
News
- Why India Needs to Build Its Own Foundational AI Model
- India Lines Up $750 million Fresh Chip Incentives
- Delhivery Turns Logistics Data Into Software, Rolls Out Maps
- AI Now Handles More Than a Third of India’s Entry-Level Work
- AI Matches Doctors in Simulated Clinical Tests
- India’s AI Talent Crunch Is Sharpest in GenAI
India needs to build its own foundational artificial intelligence model or risk becoming dependent on systems controlled by the US and China, Moneycontrol reported, citing a Bernstein report.
The brokerage said a domestic model has become a strategic necessity if India wants to retain control over the next generation of technology and remain globally competitive, according to the report.
The warning lands at a moment when India is trying to move from AI adoption to AI ownership. The government has positioned the IndiaAI Mission as the main policy vehicle for building domestic compute capacity, datasets, foundational models and AI applications, while startups and research institutions are racing to build models suited to Indian languages, use cases and public services.
The argument is no longer only about whether Indian companies can use AI. It is about whether India can shape the models, data pipelines and infrastructure on which future services, governance systems and businesses will depend.
That distinction matters because access is not the same as control. Indian companies can buy or rent access to global AI models, but those systems are trained, governed and priced outside India. Their use may also be shaped by foreign regulation, export controls, cloud contracts and the commercial priorities of companies that do not answer to Indian policy needs.
Moneycontrol reported that Bernstein’s warning was framed around the risk of India becoming dependent on US AI models, while China has shown through DeepSeek that a national technology ecosystem can produce a low-cost model capable of challenging the dominance of Western platforms.
For India, the DeepSeek comparison is uncomfortable but useful. It shows that model-building is not only a Silicon Valley game. It also raises the question of whether India’s large engineering base, public digital infrastructure and language diversity can be turned into a credible AI advantage.
The government has already moved in that direction. Under the IndiaAI Mission, New Delhi has backed indigenous foundational models, large language models and multimodal systems aligned with India’s linguistic, cultural and socio-economic diversity. The mission has also sought to expand access to compute infrastructure and India-based datasets through platforms such as AI Kosh.
During the India AI Impact Summit 2026, the government said India had already deployed 38,000 GPUs and would add another 20,000 GPUs in the coming weeks. Models developed by Sarvam AI, BharatGen, Gnani and Socket were also launched under the mission, while 12 teams were shortlisted in the first phase for indigenous foundational AI model development.
The policy push reflects a wider concern in New Delhi: the next wave of digital infrastructure may not be built around payments, identity or telecom alone, but around AI models that mediate search, work, education, health, finance, government services and software development.
That is where the sovereignty question becomes sharper. If core AI systems are foreign-built, foreign-hosted and foreign-governed, India may remain a large user market rather than a rule-setting technology power. A domestic model would not automatically solve that problem, but it could give Indian companies, researchers and public institutions more control over language coverage, data governance, pricing and deployment priorities.
The challenge is scale. Building competitive AI models requires capital, compute, talent, training data, safety testing and sustained product deployment. India has strengths in software talent and digital public infrastructure, but it still trails the US and China in frontier AI labs, advanced chips and large-scale private capital for model training.
There is also a commercial question. A national AI model cannot succeed only because it is Indian. It will need to be useful, affordable and reliable enough for enterprises, developers and government agencies to adopt it over global alternatives. That means benchmark performance, strong local language capability, data protection, enterprise support and clear accountability.
Bernstein’s warning, as reported by Moneycontrol, therefore adds market weight to a policy debate already underway. India has announced the ambition and begun assembling the building blocks. The test now is whether those blocks can produce a model ecosystem strong enough to reduce dependence without creating a protected, underused technology stack.
For India, the AI race is becoming less about copying Silicon Valley and more about deciding which parts of the technology stack must be owned at home. The answer may determine whether the country becomes an AI market, an AI maker or something in between.
