Bhavish Aggarwal, the founder of Ola, is investing $230 million into Krutrim, the AI startup he founded, as India strives to carve out a significant place in the global AI landscape, which has largely been dominated by U.S. and Chinese companies.
Aggarwal is backing the investment primarily through his family office, according to a source familiar with the matter who spoke to TechCrunch. In a post on X on Tuesday, Aggarwal revealed that Krutrim aims to raise a total of $1.15 billion by next year, with the remainder of the funds expected to come from external investors.
The announcement of this substantial investment coincides with Krutrim’s decision to make its AI models open source, alongside its ambitious plans to build what it claims will be India’s largest supercomputer, in partnership with Nvidia.
At the heart of Krutrim's AI endeavors is the recently released Krutrim-2, a 12-billion-parameter language model designed to process Indian languages. In recent sentiment analysis tests, Krutrim-2 outperformed competing models with a score of 0.95, compared to just 0.70 for others. The model also showed an 80% success rate in code-generation tasks, highlighting its potential in AI applications.
Krutrim has made several specialized models available as open source, including those for image processing, speech translation, and text search — all optimized for Indian languages. Aggarwal acknowledged the lab’s progress on X, saying, “We’re not yet at global benchmarks, but we’ve made impressive strides in just one year. By open sourcing our models, we hope to encourage the Indian AI community to collaborate and build a world-class ecosystem."
This move comes at a time when India is looking to establish itself in the competitive AI arena, which has so far been led by U.S. and Chinese firms. The recent debut of DeepSeek’s R1 “reasoning” model, developed on a reportedly modest budget, has sent ripples through the global tech industry.
India recently expressed its support for DeepSeek’s progress, announcing plans to host the Chinese AI lab's models on domestic servers. In response, Krutrim’s cloud division began offering DeepSeek’s models on Indian servers just last week.
In addition to developing powerful language models, Krutrim has also created BharatBench, an evaluation framework specifically designed to assess AI models' proficiency in Indian languages, addressing a significant gap in existing benchmarks, which typically focus on English and Chinese.
Krutrim's technical innovations include the use of a 128,000-token context window, enabling its models to handle longer texts and more intricate conversations. The performance metrics released by the startup also revealed impressive results, with Krutrim-2 scoring 0.98 in grammar correction and 0.91 in multi-turn conversations.
The funding and strategic initiatives follow the successful launch of Krutrim-1 earlier this year, a 7-billion-parameter model, which marked the introduction of India’s first large language model. The partnership with Nvidia is set to bring India’s AI capabilities to new heights, with a supercomputer launch planned for March, followed by expansion throughout the year.
Ranking
