TECH

SoftBank-Backed Billionaire to Inject $230 Million into Indian AI Startup Krutrim

Bhavish Aggarwal, the founder of Ola, is channeling $230 million into an AI startup he established as India aims to carve out a position in a sector largely led by companies from the U.S. and China.

Aggarwal is primarily funding this investment in Krutrim via his family office, according to a source who spoke to TechCrunch. In a post on X this Tuesday, Aggarwal announced that Krutrim is targeting to raise $1.15 billion by next year. He plans to source the rest of the funding from external investors, as per the source.

The funding announcement aligns with Krutrim’s recent decision to make its AI models open source, alongside plans to create what it claims will be the largest supercomputer in India in collaboration with Nvidia.

The lab has launched Krutrim-2, a language model with 12 billion parameters that has demonstrated impressive capabilities in processing Indian languages. In sentiment analysis tests released on Tuesday, Krutrim-2 achieved a score of 0.95, outperforming competing models that scored 0.70, and it boasts an 80% success rate in code generation tasks.

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

“Though we’re still far from global benchmarks, significant strides have been made in just one year,” Aggarwal, whose other projects have received backing from SoftBank, noted on X. “By open sourcing our models, we invite the entire Indian AI community to collaborate in building a world-class AI ecosystem in India.”

This initiative comes as India strives to establish a foothold in the artificial intelligence arena, which is currently dominated by U.S. and Chinese firms. Recently, the tech industry has been shaken by the launch of DeepSeek’s R1 “reasoning” model, which was developed on what is claimed to be a modest budget.

India last week lauded DeepSeek’s advancements and announced plans to host the Chinese AI lab’s large language models on domestic servers. Furthermore, Krutrim’s cloud division started offering DeepSeek on Indian servers last week.

Krutrim has also created its own evaluation framework, BharatBench, designed to assess AI models’ capabilities in Indian languages, filling a significant gap left by existing benchmarks that largely focus on English and Chinese.

The lab’s technical methodology incorporates a 128,000-token context window, enabling its systems to manage longer texts and more intricate conversations. Performance metrics from the startup revealed that Krutrim-2 achieved high ratings in grammar correction (0.98) and multi-turn conversations (0.91).

This investment follows the introduction of Krutrim-1 in January, which was India’s first large language model with 7 billion parameters. The deployment of the supercomputer in partnership with Nvidia is slated to go live in March, with expansion plans throughout the year.

Leave a Reply

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