This shift is also reflected in India’s AI stack journey. For the uninitiated, the AI stack refers to an end-to-end ecosystem spanning applications, models, compute, infrastructure and energy. Across each layer, the government and industry are pointing to operational deployments intended to extend AI benefits across sectors and regions, aiming to democratise AI for population-scale adoption.

AI applications moving from pilots to deployment

At the application layer, several use cases have moved beyond the pilot stage into sustained implementation.

In agriculture, AI-driven advisory tools are being used to guide sowing decisions, optimise input use and improve yields. State-level deployments in Andhra Pradesh and Maharashtra have reported productivity gains of up to 30 to 50 per cent, according to official notes.

Healthcare is another priority domain. AI tools are being deployed to support early detection of tuberculosis, cancer, neurological disorders and other conditions, strengthening preventive screening and diagnostic workflows within public health systems.

In education, the National Education Policy 2020 has incorporated AI literacy and applied learning through CBSE curricula, the DIKSHA digital platform and programmes such as YUVAi. The stated objective is to build foundational AI skills at scale rather than confining training to specialist institutions.

Judicial administration is also adopting AI-enabled systems. Under e-Courts Phase III, machine learning tools are being used for translation, case scheduling and workflow management, with an emphasis on improving access through vernacular languages.

Meanwhile, the India Meteorological Department is using AI for forecasting rainfall, cyclones, fog, lightning and wildfire risk. Tools such as Mausam GPT are designed to support farmers as well as disaster response agencies.

Indigenous AI models and language technologies

At the model layer, the IndiaAI Mission has extended support to develop 12 indigenous AI models targeting India-specific use cases. Startups are being offered subsidised compute access, with up to 25 per cent of compute costs offset through grants and equity participation.

The BharatGen initiative is working on India-focused foundation and multimodal models at scales ranging from billions to trillions of parameters. In parallel, IndiaAIKosh functions as a national repository for datasets, tools and models. As of December 2025, it hosts 5,722 datasets and 251 AI models contributed by 54 organisations across 20 sectors.

Language technologies remain a central priority. Bhashini, under the National Language Translation Mission, now hosts more than 350 AI models spanning speech recognition, translation, text-to-speech, optical character recognition and language detection.

Startups such as Sarvam AI are also developing large language and speech systems tailored to Indian linguistic diversity. Recently, the startup demonstrated its sovereign AI model-powered Sarvam Vision tool, which it claimed outperformed Google Gemini and ChatGPT in select but critical benchmarks related to document intelligence and speech systems.

Compute capacity and semiconductor ambitions

On compute, the IndiaAI Mission has an allocation exceeding ₹10,300 crore over five years. Its Compute Portal operates on a compute-as-a-service model, offering shared access to around 38,000 GPUs and 1,050 TPUs at subsidised rates intended to support startups and research institutions.

A separate secure national GPU cluster with 3,000 next-generation processors is under development for strategic and sovereign use cases.

Broader semiconductor ambitions are being pursued through the ₹76,000-crore India Semiconductor Mission, under which 10 projects covering fabrication and packaging have been approved. Indigenous processor programmes such as SHAKTI and VEGA are contributing to domestic capability in AI hardware.

The National Supercomputing Mission has already deployed more than 40 petaflops of computing capacity across IITs, IISERs and national laboratories. Systems including PARAM Siddhi-AI and AIRAWAT are supporting workloads such as natural language processing, weather modelling and drug discovery.

Digital infrastructure and data centre expansion

Underlying these capabilities is an expanding digital backbone. A nationwide optical fibre network supports high-speed data transfer, while 5G services are now available across all States and Union Territories, covering nearly all districts and roughly 85 per cent of the population.

India currently accounts for about 3 per cent of global data centre capacity, with installed capacity near 960 MW. Projections indicate expansion to 9.2 GW by 2030, driven by AI and cloud demand. Mumbai–Navi Mumbai remains the largest hub, followed by Bengaluru, Hyderabad, Chennai, Delhi NCR, Pune and Kolkata.

Global technology firms have also announced large investments. Microsoft has committed ₹1.5 lakh crore towards data centres and AI training initiatives, Amazon plans ₹2.9 lakh crore in cloud and AI-led infrastructure by 2030, and Google has outlined a ₹1.25 lakh crore investment for a 1 GW AI hub in Visakhapatnam.

Energy supply as a constraint and enabler

Power availability is increasingly being framed as a prerequisite for AI scale. India met a peak demand of 242.49 GW in FY 2025–26, with energy shortages reduced to 0.03 per cent. Total installed generation capacity stood at 509.7 GW as of November 2025, with non-fossil sources accounting for more than half.

Plans include 57 GW of pumped storage by 2031–32 and deployment of 43,220 MWh of battery energy storage to stabilise grids supporting data centres. The SHANTI Act further positions nuclear power — including small modular and micro-reactors — as a continuous, low-carbon energy source for compute-intensive infrastructure.