Modi government eyes India's DeepSeek in 6-10 months

The Narendra Modi government has outlined plans to develop its own large language model-based artificial intelligence system, similar to China’s DeepSeek and OpenAI’s ChatGPT, in the next six to 10 months, Union Minister of Electronics and IT Ashwini Vaishnaw said on Thursday.

The minister said the government would invite proposals to develop indigenous foundational AI models that would be aligned to Indian context, languages and culture and remove biases.

He said the AI models developed in India would be safe, secure and available at an affordable cost. He added that India has a competitive advantage in AI due to robust “compute facilities” and quality of data.

On the timeframe for development of foundational AI models in India, Vaishnaw said the outer limit is 8-10 months, while a more optimistic estimate is 4-6 months. Foundational model is a type of machine learning model that is pre-trained to perform a variety of tasks. It acts as a foundation for building more specific AI applications across different domains.

Addressing a media briefing, Vaishnaw said a common compute facility for researchers would be made operational in the next two days. “Ours is the most affordable compute facility at this point of time,” he said.

The minister claimed that India has made significant progress in the development of AI infrastructure.

The government has so far procured 18,693 graphics processing units (GPUs) under the IndiaAI Mission, which was approved by the Union Cabinet in March 2024 with a financial outlay of Rs 10,300 crore.

The minister said 10,000 GPUs are ready for immediate use, while the rest would be deployed gradually. GPU is a specialised chip designed to speed computer graphics and image processing on various devices. US-based Nvidia is the largest GPU producer, with over 80% market share globally.

For context, China’s DeepSeek-R1 model, which has grabbed global attention for accuracy and cost efficiency, has been developed using around 2,000 GPUs, while around 25,000 GPUs have been used for ChatGPT.

The centre plans to list the available GPUs in India on a portal. The GPUs will be made available for startups and researchers at much lower costs than market rates. The minister said the government will provide a 40% subsidy to students on common compute facilities

He underlined that GPU access cost globally is around $2.5 to $3 per hour (Rs 215 to 260), while it will be made available at Rs 100 per hour in India after 40% government subsidy.