Bhavish Aggarwal's AI startup, Krutrim, has launched an AI lab and committed to open sourcing its AI models, marking a significant step in India's AI landscape.

Aggarwal emphasized the goal of developing AI tailored for India, focusing on Indian languages and cultural contexts, and introduced BharatBench to evaluate AI model performance in Indic languages.

Aggarwal announced an immediate investment plan of ₹2,000 crore in Krutrim, with an additional ₹10,000 crore expected by next year, primarily sourced from his family office.

This initiative aligns with India's broader AI ambitions under the IndiaAI Mission, which focuses on democratizing computing access and enhancing data quality.

The announcement follows the rising popularity of China's DeepSeek, which has introduced a powerful open-source AI model, prompting India to accelerate its own AI development.

Union IT Minister Ashwini Vaishnaw indicated that DeepSeek models would soon be hosted on Indian servers to address data privacy concerns.

A key component of the IndiaAI Mission includes establishing a common computing facility equipped with approximately 18,693 GPUs to support startups and researchers.

The government is prioritizing the safety and ethical deployment of AI, with plans to establish an AI Safety Institute to ensure responsible development.

Krutrim-2 boasts an impressive 80% success rate in code generation tasks and can handle up to 128,000 tokens in a single session, making it suitable for complex tasks.

The competitive landscape for AI development is intensifying, particularly in response to advancements by DeepSeek, as highlighted in Aggarwal's recent announcements.

The funding will consist of both equity and debt, indicating a robust financial strategy to support Krutrim's growth.

The startup's latest model, Krutrim-2, employs a Mistral-Nemo architecture and has been trained on a diverse dataset, achieving high scores on benchmarks like IndicXTREME and IN-22.