India needs focused investment and public infrastructure to lead in AI

India's advantage lies not in emulating the US or China, but in creating a blue ocean of its own

With Aadhaar and the Unified Payments Interface (UPI), India has shown that it can solve problems at an unprecedented scale and low cost, and build digital infrastructure that the world looks to emulate. The challenge now is to replicate that achievement in AI research by identifying areas where India can lead, building the capabilities to do so, and creating solutions not just for India but for the world. 

AI or artificial intelligence research today spans 10 broad areas: Foundation models, which power general-purpose AI and are being made smaller, faster and more efficient; reasoning, which helps AI solve complex problems step by step; AI agents, which can use tools and carry out tasks independently; multimodal AI, which combines text, images, audio, video and an understanding of the physical world; AI safety, which seeks to make systems reliable, understandable and resistant to misuse; robotics, which enables machines to perceive, decide and act in the real world; AI for science, which accelerates discoveries in medicine, materials, climate and mathematics; hardware, which develops specialised chips and energy-efficient computing; data and learning methods, which help AI learn from fewer examples, generate useful training data and keep improving; and governance, which addresses fairness, privacy, transparency and regulation.   

The United States and China dominate global AI research, but with different strengths. The US leads in foundation models, reasoning, agents, safety, AI for science and hardware, supported by vast capital, computing power, and global talent. In 2025, private AI investment reached an astounding $285.9 billion. Nvidia and leading laboratories such as OpenAI, Anthropic and Google DeepMind reinforce this advantage. 

China leads in robotics and AI research output, and is rapidly closing the gap in models and chips. State support, scale and efficiency underpin its progress. DeepSeek, Alibaba’s Qwen and Moonshot’s Kimi demonstrate that powerful models can be developed at lower cost. Export controls have also accelerated China’s pursuit of chip self-reliance. Multimodal AI and data-learning methods remain fiercely contested, with both countries advancing video generation and compute-efficient training. Europe leads in governance. Its AI Act is first off the block. 

India’s position is evolving. It has genuine strengths in language AI, with AI4Bharat, BharatGen, Sarvam AI and Bhashini developing models and datasets across all 22 scheduled languages. These efforts could find markets worldwide, especially in countries where languages like Hindi, Tamil, Bengali and Punjabi are used. Applied AI is also gaining ground. Wadhwani AI has developed tuberculosis screening and cotton pest management capabilities; IIT Ropar’s Annam.AI provides local weather intelligence to farmers; and IIT Kanpur’s Airawat Research Foundation is developing AI-based air-quality systems. Hosting the AI Impact Summit, endorsed by 92 countries through the New Delhi Declaration, has strengthened India’s voice in AI governance. The IndiaAI Mission has mobilised model-development teams in multiple areas and over 45,000 graphics processing units (GPUs). Investments in data centres are growing exponentially. Yet, frontier capabilities remain limited, compute depends on imported chips, and robotics and safety research need strengthening. 

India’s digital public infrastructure offers a lesson for AI research: Rather than imitate America’s approach, India should focus on problems where its scale and diversity confer an advantage. First, multilingual, voice-first AI: Open models for underserved languages could make India a global reference point for societies across Africa, South and Southeast Asia. Second, frugal AI. India should develop efficient, affordable models that run on inexpensive devices, even offline, alongside low-power chips for phones, vehicles and farm equipment. 

Third, AI as public infrastructure. Health care, agriculture, education and land records offer opportunities to build reliable AI for environments with limited expertise, imperfect data, and strong accountability requirements. Fourth, evaluation and trust. India can develop multilingual benchmarks, test for social biases, detect deepfakes and establish safety standards relevant to the Global South. Achieving this will require substantially deeper research capabilities. 

The following six steps may turn these priorities into reality. First, create a dedicated institution. A National AI Development Organisation, structured as a non-govt public entity, with 26 per cent government ownership, 26 per cent held by State Bank of India and Life Insurance Corporation, 30 per cent by the IT industry, and the balance by philanthropic and academic institutions, could provide the focused leadership that National Payments Corporation of India brought to digital payments. It could build critical AI research and deployment infrastructure. 

Second, provide ₹25,000 crore -30,000 crore annual funding for a decade concentrated in selected priorities rather than spread thinly. Third, publicly-funded models, datasets and benchmarks should be openly available. The Data Empowerment and Protection Architecture , which enables secure, consent-based AI training should be rolled out on a mission-mode. Formally adopting it could turn AIKosh and Bhashini into trusted data commons for health, agriculture, education and speech. Fourth, make government the first customer. It can create demand and help promising Indian AI solutions scale. Fifth, invest in talent. A thousand research chairs, several thousand doctoral fellowships modelled on the Vishveshwariah PhD scheme of 2014, and efforts to attract Indian researchers abroad could strengthen research capacity.  Sixth, expand computing and chip design. Scale shared GPU access, fund national supercomputing, and support energy-efficient AI chips through Semicon India, combining design grants, fabrication access and assured government purchases. 

India must not repeat its semiconductor experience in AI. Indian talent designed world-class chips, but the intellectual property (IP) belonged largely to global majors. India must urgently create and scale its own IP, for India and the world. India should not race to build AI that merely matches the world’s best. That is a red ocean, crowded with giants. Its opportunity lies in a blue ocean of its own: AI that speaks every language and runs on the billions of devices in people’s hands.

The writer is former Defence Secretary and Chairman, UPSC. The views are personal