The Office of the Principal Scientific Adviser (PSA) has released a white paper detailing India's strategic pivot toward building indigenous Foundation Models (FMs) to ensure digital sovereignty and cultural inclusivity. Central to this vision is the IndiaAI Mission, approved with a budget of ₹10,371.92 crore, which aims to reduce reliance on foreign AI models that often under-represent Indian languages and contexts.
The strategy leverages a public-private collaboration model, supporting projects like BharatGen, Sarvam AI, and Gnani AI to develop large and small language models specifically for agriculture, healthcare, and governance. This ecosystem is powered by the IndiaAI Compute Portal, which provides startups and researchers subsidized access to over 38,000 GPUs, and the AIKosh platform, hosting more than 10,000 datasets.
Key Pillars of India's Indigenous AI Strategy
Strategic Autonomy: Developing domestic FMs to mitigate biases in foreign models and ensure AI outputs reflect India’s linguistic and cultural diversity.
Massive Compute Infrastructure: Shared access to high-end resources via a central portal, significantly lowering the entry barrier for AI startups and academia.
India-Centric Data (AIKosh): A repository of 10,021+ datasets across 20 sectors, enabling the training of models on high-fidelity, locally relevant information.
Robust Governance: The India AI Governance Guidelines (2025) and the DPDP Act (2023) provide a framework for accountability, data privacy, and purpose limitation.
Localized Benchmarking: Implementation of evaluation tools like Indic-Bias, MILU, and EKA-Eval to test FM performance on Indian accents, dialects, and cultural validity.
IP & Innovation Balance: A proposed hybrid model by DPIIT to balance AI training needs with copyright protection through blanket licenses and commercial royalties.
What is a "Foundation Model" (FM)? A Foundation Model is a large-scale AI system trained on vast amounts of data that can be adapted (fine-tuned) to a wide range of downstream tasks and sectors. Unlike traditional AI designed for a single purpose, FMs play a role as a "base" technology that powers diverse applications, from chatbots to scientific discovery tools. In the Indian context, indigenous FMs are supported by the goal of creating "multimodal" systems capable of understanding speech, text, and images across multiple Indian languages. This foundational approach reflects growth in India’s deep-tech capabilities, allowing the same underlying model to drive innovations in varied fields like personalised education and climate-resilient farming.
Policy Relevance: Anchoring AI in National Priorities
Scaling Linguistic Inclusivity: Developing models like BharatGen ensures that AI services are accessible to non-English speakers, directly addressing the digital divide.
Internalising Data Sovereignty: Using the AIKosh platform to train indigenous models plays a role in keeping sensitive Indian data within domestic jurisdictional boundaries.
Bypassing High Entry Costs: Subsidised GPU access is supported by the need to foster a competitive startup ecosystem that is not stifled by the prohibitive cost of private compute.
Supporting Ethical AI: The use of Indic-centric benchmarks contributes to identifying and removing western-centric biases, ensuring AI tools are fair and culturally valid for Indian citizens.
Follow the Full White Paper Here: Office of the PSA: Advancing Indigenous Foundation Models
