New Delhi: As the global AI landscape evolves, Bengaluru-based start-up Sarvam is stepping into the spotlight, tasked with creating India’s first indigenous large language model (LLM) under the government’s ambitious IndiaAI Mission. This initiative comes in the wake of DeepSeek, China’s disruptive AI model that has made waves for its low-cost development and open-source approach.
Sarvam, chosen from 67 applicants, is receiving substantial government support, including access to 4,000 GPUs for six months, to build its model from scratch. The company’s AI model is designed for advanced reasoning, fluency in Indian languages, and optimized for voice—a critical step for India to assert its AI leadership in a rapidly changing global landscape.
While DeepSeek, China’s emerging AI contender, has already made a significant mark with its cost-effective foundational model—said to rival top US models in accuracy but trained on less powerful GPUs—Sarvam’s model aims to deliver India-specific innovations. Sarvam’s LLM will feature 70 billion parameters, positioning it as a robust alternative in the growing AI race. The model is not expected to be open-source, unlike DeepSeek, but will be specially tuned for Indian languages, setting it apart in terms of localization.
Sarvam will focus on building its model entirely within India, using local infrastructure and talent—a strategic move to maintain AI sovereignty. This contrasts with DeepSeek’s reliance on open-source accessibility and global competitiveness, which has already had a noticeable impact on the stock market, particularly affecting Nvidia’s shares due to DeepSeek’s ability to use lower-cost GPUs effectively.
In addition to the main LLM, Sarvam is developing three specific variants; Sarvam-Large for high-end tasks requiring advanced reasoning, Sarvam-Small for real-time interactive applications, and Sarvam-Edge for compact, on-device AI tasks.
The aim is not only to rival international AI models but also to ensure India’s strategic autonomy in the AI space, limiting dependence on foreign technology while driving innovation within the country.
As part of the IndiaAI Mission, the government has already selected 10 companies to provide over 18,600 GPUs, surpassing initial procurement goals. This new infrastructure will lay the foundation for Sarvam’s ambitious AI model and will contribute to India’s growing presence in the global AI ecosystem, competing directly with disruptive models like DeepSeek from China.
In this context, Sarvam’s AI development presents a unique opportunity for India to establish a distinct AI identity, integrating local language capabilities and cultural relevance into cutting-edge technology, all while reinforcing national security and data sovereignty.
BI Bureau
