Bhavish Aggarwal’s AI startup Krutrim brings DeepSeek R1 to Indian developers

Krutrim, an artificial intelligence (AI) startup co-founded by Ola founder Bhavish Aggarwal, has unveiled DeepSeek R1 671B in India at a price that is creating buzz among developers. 

The company is offering access to the powerful open-source AI model for just Rs 1 per million tokens throughout February, in an effort to encourage widespread AI adoption and innovation in the country.

Caution and opportunity

Aggarwal highlighted the need to be cautious with the DeepSeek app itself but stressed that the open-source version of DeepSeek R1 can be safely deployed on Indian servers to accelerate the nation’s AI progress.

The model, hosted on Nvidia’s H100 graphics processing units, is now available through Krutrim AI Cloud. With this move, Krutrim becomes the first in the world to deploy DeepSeek-R1 671B on H100 GPUs, and it hopes to drive AI adoption across Indian businesses and among local developers.

Comparisons with other models

At the promotional rate of Rs 1 per million tokens for February, Krutrim’s DeepSeek R1 deployment is far more affordable than other large language models. 

By contrast, developers pay around Rs 16.60 per million tokens for Llama-3-8B-Instruct, Rs 66.40 per million tokens for Google’s Gemma-27B and nearly Rs 16.60 per million tokens for Hugging Face’s M4/idefics2-8B. 

Aggarwal has described this pricing as a strategic move to encourage experimentation and bring more Indian users into the AI ecosystem.

Krutrim’s supercomputing plans

Krutrim is also planning to deploy Nvidia’s GB200 GPUs by March 2025 to create what Aggarwal calls “the largest supercomputer in India by the end of the year.” This computing infrastructure is set to help train and manage AI models on a scale rarely seen in the country.

Earlier, Aggarwal had announced that Krutrim would open source much of the company’s work in 2024, aiming to build a robust AI community in India that can collectively compete with global leaders.

New AI lab and funding

In addition to its deployment of DeepSeek R1, Krutrim has announced the launch of Krutrim AI Lab, a ₹2,000-crore initiative dedicated to AI research and open-source model development. The lab’s investment is expected to rise to Rs 10,000 crore by next year. 

Aggarwal hopes this large funding commitment will bring together India’s AI community, enabling developers to work collaboratively on overcoming the challenges of data scarcity and cultural context in Indian languages.

Focus on India

Launched in April 2023, Krutrim—whose name comes from a Sanskrit word meaning “artificial”—had already claimed unicorn status by January last year. 

Aggarwal co-founded the venture with Krishnamurthy Venugopala Tenneti, a board member of ANI Technologies, which owns Ola and Ola Electric.

Krutrim’s open-source approach, its plans for a homegrown supercomputer and the availability of DeepSeek R1 at such a low cost are all part of a larger vision.

Aggarwal believes India’s AI community will benefit from local hosting of powerful models, and he aims to foster a collaborative environment that ensures future AI developments cater to Indian users’ languages and cultural context.