Should India build its own AI model

Source: The post Should India build its own AI model has been created, based on the article “Should India build a sovereign, foundational AI model?” published in “The Hindu” on 7th February 2025.

UPSC Syllabus Topic: GS Paper3-Science and Technology-indigenization of technology and

developing new technology And Awareness in the fields of IT, Computers

Context: The article discusses whether India should build its own AI foundation model. Experts debate if it is necessary for sovereignty or just national pride. They highlight high costs, research challenges, and the need for smart investments in AI infrastructure and applications.

For detailed information on On Building Sovereign AI in India – India’s challenge read this article here

Does India need to develop its own foundational AI model?

1. Experts debate whether India should build a sovereign AI model. Some argue that sovereignty concerns, such as U.S. sanctions on AI chips, make it necessary.
2. Others believe that open-source models like DeepSeek R1 reduce this need. Instead of focusing on sovereignty, India should ensure it has the capability to build and utilize AI models effectively.

Can India afford to build an AI model, and does it have the technology?

1. Developing a foundational AI model requires huge investments and advanced technology. DeepSeek V3 cost $5.6 million, while Big Tech firms spend $80 billion annually on AI infrastructure.
2. India lacks semiconductor manufacturing and has no contracts with TSMC.
3. Unlike Huawei’s 910C chips, India does not have high-performance GPUs needed for AI model training.
4. Additionally, the Indian AI market is smaller than the U.S., making it harder to recover the investment.

What is the best AI strategy for India?

1. Focus on AI applications: Instead of building a costly foundational AI model, India should develop AI solutions for governance, businesses, and Indian languages. (Example: AI for Bharat is training IndicTrans2 and a text-to-speech system for Indian languages.)
2. Invest in R&D and infrastructure: The IndiaAI Mission is providing subsidized GPUs to startups and academia to lower AI development costs. However, India’s public procurement system lacks error tolerance, making private investment and research critical.
3. Leverage open-source models: DeepSeek R1 and other open-source models reduce dependency on proprietary AI. If the U.S. applies sanctions, India can “fork” existing open-source AI models and modify them.
4. Be realistic about AI investments: DeepSeek V3’s training cost $5.6 million, and Big Tech spends $80 billion annually on AI infrastructure. India’s limited AI budget is less than the cost of training Meta’s Llama 4 model, so resources must be spent wisely.
5. Prioritize language and industry needs: Developing AI models tailored to Indian language processing and enterprise applications will offer better returns. Competing directly with ChatGPT or DeepSeek is impractical.

Question for practice:

Examine whether India should prioritize building its own foundational AI model or focus on AI applications and infrastructure development.