India Releases White Paper on Advancing Indigenous Foundation Models

On March 13, 2026, the Office of the Principal Scientific Adviser to the Government of India released a white paper titled 'Advancing Indigenous Foundation Models,' detailing strategies to build sovereign AI systems trained on India-specific data across 22 scheduled languages and key sectors.

Tech Insights Reporter Mar 13, 2026 5 min read New Delhi

TLDR

The Office of the Principal Scientific Adviser released the white paper "Advancing Indigenous Foundation Models" on March 13, 2026. It maps ongoing public-private efforts to develop large and small language models, voice, and multimodal systems optimized for Indian languages, culture, and priority use cases under the IndiaAI Mission.

India's Strategy for Indigenous Foundation Models

Foundation models underpin many modern AI applications because they can be adapted across tasks. The white paper notes that most widely used models today are developed abroad on datasets that under-represent India's linguistic and cultural diversity. This creates risks of bias and limits local relevance in areas such as healthcare, education, agriculture, and governance.

India's approach centers on building indigenous capability through the IndiaAI Mission, approved by Cabinet in March 2024 with an outlay of ₹10,371.92 crore over five years and led by the Ministry of Electronics and Information Technology (MeitY). Two dedicated pillars — the India AI Compute Portal and the AI-Kosh platform — provide shared access to compute and high-quality datasets to lower barriers for startups, academia, and public sector teams.

Call for Proposals and Selected Initiatives

In January 2025 the IndiaAI Mission issued a Call for Proposals for foundational AI models. It received 506 proposals by April 2025. Expert evaluation led to selections in phases:

Phase 1: Sarvam AI, Soket AI, Gnani AI, and Gan AI — focusing on multilingual text, voice AI, and advanced text-to-speech.
Phase 2 (announced September 2025): Eight additional initiatives to develop large and small language models spanning all 22 scheduled Indian languages.

Models and prototypes already emerging include Sarvam AI's Sarvam-105B (trained from scratch and optimized for Indic performance), Gnani.ai's Inya VoiceOS (voice-to-voice processing supporting over 15 languages), Fractal's Vaidya 2.0 (healthcare reasoning and diagnostics), Tech Mahindra's Project Indus with NVIDIA (Hindi-first for education), Zoho's Zia LLM for enterprise workflows, and CoRover's BharatGPT (multilingual conversational model in 12 languages). Academic efforts such as BharatGen at IIT Bombay have delivered Param-1 (text), Shrutam (speech), Sooktam (text-to-speech), and Patram (document understanding).

The white paper stresses a portfolio approach: frontier-scale LLMs alongside smaller, efficient SLMs tailored for domain-specific tasks with lower compute and energy demands. These can be fine-tuned for agriculture, health, education, MSMEs, climate, and urban governance while leveraging shared foundational layers.

Governance and Data Priorities

The document situates model development within India's existing framework, including the India AI Governance Guidelines (2025), the Digital Personal Data Protection Act, and the IT (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021. It calls for India-centric benchmarks, balanced treatment of intellectual property, and continued investment in non-personal datasets via AI-Kosh to support responsible innovation.

Selected models are expected to become available through the AI-Kosh platform for broader use by the startup and research community.

Why this story matters

The March 13 white paper provides a clear public snapshot of India's multi-pronged push for technological sovereignty in AI. By prioritizing models that reflect local languages and contexts, combined with shared national infrastructure, the strategy aims to expand access beyond large players and reduce dependence on foreign systems. This has direct implications for equitable AI adoption across India's diverse population and economy.

Sources

Office of the Principal Scientific Adviser to the Government of India — "Whitepaper on Advancing Indigenous Foundation Models," published March 13, 2026. https://www.psa.gov.in/ai-mission-initiatives and PDF at https://psa.gov.in/CMS/web/sites/default/files/publication/WP-Foundation%20Model%20New%20Edited%20version%20%281%29.pdf
CNBC TV18 — "India developing multiple model-building efforts across text and speech: White Paper on AI policy" (March 14, 2026). https://www.cnbctv18.com/technology/india-developing-multiple-model-building-efforts-across-text-and-speech-white-paper-on-ai-policy-ws-l-19868262.htm
Additional reporting in Communications Today and policy summaries referencing the March 13 release.

Featured Image Alt Text

Document cover or official graphic representing India's white paper on advancing indigenous AI foundation models, with icons for language diversity and compute infrastructure