Sarvam AI and Krutrim: India's Homegrown LLM Builders

Imagine India's massive linguistic diversity, dozens of major languages and hundreds of dialects, being poorly served by large language models trained predominantly on English and other major global languages, creating genuine space for homegrown AI startups specifically focused on building models that actually understand and generate fluent Indian-language text, exactly the mission behind Sarvam AI and Krutrim, two of India's most prominent large language model builders.

Sarvam AI, having raised $53 million in its Series A from investors including Lightspeed, Peak XV and Khosla Ventures and reached unicorn status at a $1.5 billion valuation, represents a dedicated foundational model research effort, while Krutrim, built by ride-hailing company Ola and recognised as India's first AI unicorn, has taken a broader approach, building both its own models and Krutrim Cloud, a GPU cloud service offering H100 computing time to Indian enterprises alongside investment in inference infrastructure specifically for real-time Indian-language applications.

Both companies operate directly within the GPU access constraints covered under the chip import restrictions discussion elsewhere on this site, Sarvam's own compute allocation under the IndiaAI Mission illustrates how even well-funded, government-prioritised AI startups remain genuinely bottlenecked by chip availability, meaning building competitive foundational models in India requires navigating both the technical challenge of the models themselves and the harder infrastructure challenge of simply accessing enough computing power to train them.

This homegrown model-building effort connects directly to the Sovereign AI Models and DPDP Act data localisation discussions covered elsewhere on this site, models trained specifically on Indian languages and cultural context, and ideally run on infrastructure physically located within India, address both a genuine capability gap, global models' comparatively weaker performance on Indian languages, and a strategic data sovereignty concern, reducing dependence on foreign AI infrastructure for applications serving Indian citizens and businesses.