China’s AI breakthrough challenges India to develop its own foundation model
India has long debated whether it should build its own AI foundation models. Aadhaar architect Nandan Nilekani previously argued that India should focus on AI applications rather than investing in expensive foundation models.
However, China’s DeepSeek has now trained a frontier AI model with just $5.6 million and a modest number of graphics processors, weakening the argument against an Indic foundation model, reported timesofindia.indiatimes.com.
Anil Pawar, chief AI officer at Yotta Data Services, believes an Indic foundation model would enable language and culture-specific solutions catering to India’s diverse population.
Experts argue that Western large language models often fail to understand Indian contexts, producing inaccurate or irrelevant responses to cultural prompts.
Pawar emphasizes that while India is data-rich, the real challenge lies in optimizing algorithms for efficiency.
DeepSeek’s success shows that reasoning-based techniques can significantly reduce costs.
Gautam Singh of WNS Analytics highlights that innovations in hardware optimization, streamlined architectures, and high-quality datasets could help India achieve similar AI breakthroughs
Linguistic complexity and data scarcity pose major challenges.
Building an Indic LLM comes with significant hurdles. Samiksha Mishra, director of AI at R Systems, notes that India’s linguistic diversity makes AI development difficult.
She points out that an Indian LLM must accommodate over 22 official languages and up to 1,600 dialects, requiring extensive data curation and infrastructure investment.
Developing an Indic AI model is essential for India’s technological and cultural growth
Mishra adds that Indic scripts are morphologically rich and feature complex ligatures, making model training more challenging.
Ensuring linguistic inclusivity is crucial so that dominant languages do not overshadow smaller but culturally significant ones.
Open-source frameworks could help bridge this gap.
Despite growing interest in open-source AI, India has yet to release a homegrown LLM of significance Proprietary efforts like CoRover’s BharatGPT remain limited in scope.
CoRover CEO Ankush Sabharwal stresses the need for better data aggregation and ownership to build robust AI models that protect privacy while embracing India’s linguistic diversity.
A large-scale initiative is required to push India ahead in AI
Experts argue that India’s traditional focus on operational efficiency rather than AI research could slow its progress Publicis. Sapient’s Tulika Sanghi believes India must invest in innovation rather than just scalability to compete globally.
Sanjay Koppikar, co-founder of EvoluteIQ, calls for a structured, large-scale initiative similar to Aadhaar.
He notes that while projects like Bhashini, Sarvam, and Krutrim show promise, they remain fragmented and underfunded.
Without a unified vision and dedicated funding, India risks falling behind in AI innovation.
***********************************************************
Readers
These are extraordinary times. All of us have to rely on high-impact, trustworthy journalism. And this is especially true of the Indian Diaspora. Members of the Indian community overseas cannot be fed with inaccurate news.
Pravasi Samwad is a venture that has no shareholders. It is the result of an impassioned initiative of a handful of Indian journalists spread around the world. We have taken a small step forward with the pledge to provide news with accuracy, free from political and commercial influence. Our aim is to keep you, our readers, informed about developments at ‘home’ and across the world that affect you.
Please help us to keep our journalism independent and free.
In these difficult times, running a news website requires finances. While every contribution, big or small, will make a difference, we request our readers to put us in touch with advertisers worldwide. It will be a great help.
For more information: pravasisamwad00@gmail.com