TECH

Billionaire Backed by SoftBank Invests $230 Million in Indian AI Startup Krutrim

Bhavish Aggarwal, the founder of Ola, is channeling $230 million into an AI startup he established, as India aims to position itself in a field largely dominated by U.S. and Chinese firms.

The bulk of Aggarwal’s investment in Krutrim is sourced from his family office, as per a source who communicated with TechCrunch. On X this Tuesday, he revealed that Krutrim is looking to secure $1.15 billion by next year, with the balance anticipated to come from outside investors, according to the source.

This funding announcement aligns with Krutrim’s recent decision to open-source its AI models, while also planning to create what it claims will be India’s largest supercomputer in partnership with Nvidia.

The lab has introduced Krutrim-2, a language model with 12 billion parameters, demonstrating impressive capabilities in processing Indian languages. Based on sentiment analysis assessments unveiled on Tuesday, Krutrim-2 attained a score of 0.95, outshining rival models that recorded scores of 0.70, and achieving an 80% success rate in coding assignments.

Moreover, the lab has made several specialized models available as open source, including those for image processing, speech translation, and text search, tailored specifically for Indian languages.

“Though we have considerable progress to make to reach global benchmarks, we’ve accomplished a lot in just one year,” Aggarwal, whose ventures have drawn investments from SoftBank, stated on X. “By open-sourcing our models, we invite the entire Indian AI community to collaborate with us in developing a world-class AI ecosystem in India.”

This effort comes as India strives to secure its position in the artificial intelligence arena, which is currently led by companies from the U.S. and China. The tech sector has recently been invigorated by DeepSeek’s introduction of its R1 “reasoning” model, reportedly developed on a limited budget.

Last week, India commended DeepSeek’s advancements and expressed plans to host the Chinese AI lab’s large language models on local servers. Concurrently, Krutrim’s cloud division began providing DeepSeek on Indian servers during the same week.

Krutrim has also developed its own evaluation system, BharatBench, designed to assess AI models’ effectiveness in Indian languages, thereby filling a critical void left by current benchmarks that focus mainly on English and Chinese.

The lab’s technical approach incorporates a 128,000-token context window, enabling its systems to manage longer pieces of text and more intricate interactions. Performance indicators from the startup show that Krutrim-2 scored highly in grammar correction (0.98) and multi-turn dialogues (0.91).

This investment follows the unveiling of Krutrim-1 in January, marking India’s inaugural large language model featuring 7 billion parameters. The deployment of the supercomputer in collaboration with Nvidia is slated to begin in March, with expansion plans set for throughout the year.

Leave a Reply

Your email address will not be published. Required fields are marked *