India Pushes for Self-Reliant AI
Indian startup Sarvam AI has open-sourced new AI models supporting 22 Indian languages, built using government supercomputers. The move comes as Delhi hosts a major AI Impact Summit, signaling a major push towards building a sovereign AI infrastructure.
This push is part of the government's broader IndiaAI Mission, a five-year plan with an outlay of ₹10,371.92 crore (approximately $1.2 billion) aimed at building a robust domestic AI ecosystem. The mission's goal is to foster technological self-reliance and ensure AI development is tailored to India's unique social and economic landscape. Sarvam AI's flagship open-source model, Sarvam 105B, utilizes a "Mixture-of-Experts" (MoE) architecture. This means that while the model has 105 billion parameters, only about 10.3 billion are active for any given task, making it highly efficient without sacrificing performance in complex reasoning, mathematics, and coding. The Bengaluru-based startup, founded by former AI4Bharat researchers, secured $41 million in a funding round led by Lightspeed Venture Partners, with participation from Peak XV Partners and Khosla Ventures. The compute power for these models comes from the IndiaAI Compute Capacity pillar, which aims to establish a large-scale AI infrastructure of over 10,000 Graphics Processing Units (GPUs). This initiative provides startups and researchers with subsidized access to high-performance computing, a resource traditionally controlled by a few global players. More than 38,000 GPUs have already been onboarded for this common facility. A significant infrastructure boost was announced at the AI Impact Summit: a partnership with UAE-based G42 and US firm Cerebras to build a national-scale AI supercomputer in India with 8 exaflops of computing power. This system will operate under Indian governance frameworks, ensuring all data remains within national jurisdiction, a critical component of the sovereign AI strategy. The Indian artificial intelligence market is projected to grow significantly, with one forecast expecting it to reach over $130 billion by 2032, up from about $7.6 billion in 2024. Another projection estimates the market could generate revenue of over $325 billion by 2033, growing at a CAGR of 38.1% from 2026. Sarvam AI is one of twelve organizations, alongside others like Soket AI and Gnani AI, selected by the government to develop indigenous foundational models. This is part of the IndiaAI Mission's strategy to create Large Multimodal Models trained on Indian datasets, reflecting the nation's linguistic and cultural diversity.
