Should India build a sovereign, foundational AI model?

• OpenAI CEO Sam Altman’s remarks on the cost of training foundational AI models in India sparked debate on whether Indian firms should invest in such technology.

• The issue revolves around sovereignty, national pride, and financial constraints.

• Pranesh Prakash and Tanuj Bhojwani discuss the necessity of India working towards a foundational model, as announced by the IT Ministry’s IndiaAI Mission.

• Prakash believes India should focus on foundation models, not for sovereignty. He emphasizes the importance of having people capable of building foundation models and applications.

• Bhojwani suggests that India should not worry about sanctions and export controls over proprietary models, as India has many open weights and open source models.

• Bhojwani also emphasizes the need for GPUs to run a model, as DeepSeek uses Huawei’s 910C chips.

• Bhojwani emphasizes the importance of seeing returns on investment and ensuring the investment is proportionate to the end goal.

• The debate remains on whether India needs to build a sovereign foundational AI model for its local needs and if the emergence of DeepSeek at low cost makes it more attainable.

AI Investment in India: Challenges and Opportunities

• The Indian AI market is largely U.S.-dominated, with 60% of the global market cap being in the U.S.

• The Indian enterprise market that will purchase AI is smaller, making it challenging to create an Indian model with local language content.

• The government needs to create an institute with autonomy and spending power to succeed, as 90% of the spending will go to failed experiments.

• There is a strong enthusiasm among young developers, who are learning to use six different development tools from the very beginning.

• The government should promote responsible usage of these technologies, including careful studies of regulations.

• The IndiaAI Mission under the IT Ministry is making GPU clusters available to startups and academia at subsidised rates.

• The government is not taking a concentrated bet, but is thinning out sparse resources.

• AI for Bharat is using GPUs to train IndicTrans2, a text-to-speech system for Indian languages.

• The government should acknowledge its limited resources and consider the wisest way to spend them.