India’s AI policy highlights smaller language models

India’s artificial intelligence roadmap is placing strong emphasis on smaller, specialized AI models designed for practical deployment across sectors like agriculture, healthcare and education. A government white paper highlights how these systems, combined with large foundation models, could shape a more inclusive and cost-efficient AI ecosystem in the country.

India’s artificial intelligence strategy is increasingly focusing on smaller, sector-specific AI systems, even as large global AI models continue to dominate headlines.

A white paper titled “Advancing Indigenous Foundation Models”, released by the Office of the Principal Scientific Adviser to the Government of India on March 13, outlines the country’s approach to developing AI technologies tailored to local needs.

The document forms part of the government’s AI Policy White Paper Series, which aims to guide the growth of India’s domestic artificial intelligence ecosystem while aligning technology development with national priorities, societal values and regulatory frameworks.

Indigenous Foundation Models a Strategic Priority

The white paper identifies the development of indigenous foundation models as a key priority. These large AI systems are trained on extensive datasets such as text, images, audio and video, enabling capabilities like translation, summarisation, classification and question answering.

Building such models using India-specific datasets and governance frameworks is seen as critical to ensuring transparency, inclusivity and alignment with national interests.

However, the document also highlights the importance of Small Language Models (SLMs) that are easier to deploy and more cost-efficient for practical applications.

Why Smaller AI Models Matter

According to the white paper, smaller AI systems can be highly effective in sectors such as:

Agriculture

Healthcare

Education

Micro, small and medium enterprises (MSMEs)

These models require far less computing power and energy compared with massive frontier models, making them more suitable for large-scale deployment in developing economies.

The strategy does not replace large models but instead positions small and large AI systems as complementary layers within the broader ecosystem.

Growing Ecosystem of Indigenous AI Projects

Several Indian initiatives are already developing models focused on local languages and applications.

Sarvam AI has introduced large language models optimized for Indic languages, including Sarvam-105B.

BharatGen, led by Indian Institute of Technology Bombay, has created AI systems such as Param-1 for text processing and Shrutam for speech recognition.

Zoho has developed its enterprise AI model called Zia LLM for business workflows.

CoRover.ai launched BharatGPT, a multilingual conversational AI platform trained on Indian datasets.

Voice-based AI is also expanding, with systems designed to process speech directly across multiple languages.

Infrastructure Supporting India’s AI Push

To support domestic AI development, the government launched the IndiaAI Mission, approved by the Union Cabinet in March 2024 with a budget of ₹10,371.92 crore over five years.

The initiative focuses on building a multi-layered AI ecosystem that includes:

Shared compute infrastructure

Large national datasets

Language technology tools

Indigenous foundation models

Through the IndiaAI Compute Portal, developers can access large-scale GPU resources. More than 38,000 GPUs have already been onboarded, with subsidised computing access priced at around ₹65 per hour.

Language and Data Infrastructure

India’s AI ecosystem also includes platforms aimed at supporting linguistic diversity.

The Bhashini initiative is developing benchmarks and evaluation tools for speech and language AI across India’s languages.

Meanwhile, AIKosh serves as a national repository hosting datasets, models and resources required for training foundation models.

Together, these initiatives aim to support AI systems across all 22 scheduled Indian languages and multiple regional dialects.

Expanding Model Development Programs

Under the IndiaAI Mission, the government has invited researchers, startups and companies to develop AI systems tailored to India’s needs.

A call for proposals in 2025 received over 500 submissions, with multiple organizations selected to build multilingual language models, speech technologies and AI-driven voice systems.

The projects span a range of organizations including technology companies, AI startups and academic research groups, forming a pipeline that connects datasets, infrastructure and deployable AI models.

The white paper also proposes exploring new governance frameworks for AI training data, including a possible licensing system that allows developers to train models on lawfully accessed content while paying royalties when commercialised.

Overall, India’s AI strategy aims to create a balanced ecosystem where large foundation models and smaller, specialized systems work together to support economic growth, public services and technological innovation.