At the India AI Impact Summit 2026, Sarvam AI unveiled two indigenous large language models—Sarvam-3...
The AMW Read
Updates the Foundation Model player map with a major regional contender and signals a structural shift toward sovereign AI ecosystems in India.
NoveltySignificance
Foundation Models · Player MapGeopolitics
At the India AI Impact Summit 2026, Sarvam AI unveiled two indigenous large language models—Sarvam-30B trained on 16 trillion tokens and Sarvam-105B with a 128K context window—both designed to support all 22 scheduled Indian languages and built using Mixture-of-Experts architecture for cost-efficient inference. The company, founded by Vivek Raghavan and Pratyush Kumar in 2023 with $54M in funding from Lightspeed and Peak XV, is open-sourcing these models to power India's sovereign AI ecosystem across government and enterprise deployments. This milestone signals India's strategic shift from consuming global AI to building indigenous alternatives optimized for local cost structures and multilingual scale.
