What Is Open-Source Today May Not Remain So Tomorrow: Vaishnaw On India’s LLM Efforts

India is developing domestic large language models (LLMs) as the artificial intelligence models that are open-source currently may not remain so in the future, just as Microsoft-backed OpenAI flipped its model to become closed, Union Minister Ashwini Vaishnaw said in Delhi on March 18.

Vaishnaw also highlighted India's successful implementation of 5G and semiconductor initiatives as potential models for accelerating AI development in the country.

Speaking at the 10th edition of ‘The Raisina Dialogue’, the Minister also stressed on talent development as a critical area where India has an edge in AI journey. 

“We've already made significant progress by inaugurating 14,000 GPUs as a common compute facility. Two critical aspects of our AI journey are access to datasets and infrastructure. We've set up a unified compute structure called AI Kosh, and just this morning, we signed an MoU with the India AI Mission and the Indian Parliament”, said Vaishnaw. 

He added, “Given the vast, multilingual datasets that the Parliament has accumulated over time, this partnership will serve as a valuable resource for training AI models. Additionally, other data sources like Doordarshan and All India Radio will further enrich our dataset pool.”

India is rapidly building a strong AI computing and semiconductor infrastructure to support its growing digital economy. With the approval of the IndiaAI Mission in 2024, the government allocated Rs 10,300 crore over five years to strengthen AI capabilities. 

A key focus of this mission is the development of a high-end common computing facility equipped with 18,693 Graphics Processing Units (GPUs), making it one of the most extensive AI compute infrastructures globally. 

“Developing our own GPU chip is a top priority, and we're actively collaborating with industry experts, startups, and professors to explore the best options. We hope to finalize our approach soon,” informed the IT Minister. 

Echoing concerns raised by other panelists about the limitations of relying on open-source systems for the long term, Vaishnaw emphasised India's goal to develop its own large language model (LLM). 

He noted that this move aims to reduce dependency on potentially unstable open-source options, citing OpenAI's evolving approach as an example.