India to Develop Local Language AI Foundation Model Amid Global AI Race

AI (model) of the tiger: India to back local language AI foundation model, RFP to be issued soon

India will back an industry-academia consortium to build a local language artificial intelligence (AI) foundation model and will soon issue request for proposal (RFP) to further the effort, a senior official told ET. The Centre's move, which follows the US government restricting export of AI compute infrastructure and limiting use of language models that are currently dominated by American technology firms, will place India in a select group of nations including the US and China where local companies are building prominent models. Foundation models are artificial intelligence (AI) models trained on vast repositories of data, which can be used as the base for other applications. The debate over whether India should have its own foundational model or just work on open-source models "was going on for a long time, but it's no longer an either-or debate anymore. We need to do both", said the official cited above. The government will form a technical committee which will reach out to startups and academic institutions to access the technical and financial support required for such an initiative. Industry backing "It has to be a consortium, we can't put all our eggs in one basket," sources said, noting that the model based on Indian languages and datasets will (subsequently) be open sourced for the larger industry to ride on. "The action by the US kind of pushes us necessarily in that direction," the person added. The outgoing Joe Biden administration passed an executive order on January 13 which restricted nations like India from importing more than 50,000 GPUs along with imposing limitations on use of AI models. On Tuesday, the new US president Donald Trump also backed an Open AI-led industry collaboration to invest $500 billion towards AI initiatives in the US. "As AI becomes more powerful, countries are likely to impose restrictions on sharing these technologies, just as we've seen with AI chips. India cannot afford to be left behind," said Srikanth Velamakanni, co-founder of Fractal.ai. He said the real reason India must build its own foundational models is to advance AGI or Artificial General Intelligence, which aims to create machines with human-like intelligence. The Centre's move comes amidst calls by Indian-origin technology leaders such as Aravind Srinivas, co-founder of one of the most valuable AI startups Perplexity for India to not "fall into the trap" of thinking that (foundational) models need a lot of money to train. The cost of training and housing of such models is steep, running into millions of dollars while also requiring high-powered compute or GPUs that are an expensive and scarce resource currently. However, the recent launch of open-source AI reasoning model by a Chinese AI startup-DeepSeek-has raised questions on that theory. The Chinese free, and open-source model is arguably rated as better than the ones built by American majors Meta and OpenAI and was developed at a significantly lower cost. "I hope India changes its stance from wanting to reuse models from open source and instead trying to build muscle to train their models that are not just good for Indic languages but are globally competitive on all benchmarks," Srinivas wrote on X on Tuesday. He followed it up on Wednesday with an offer to "invest $1 million personally and 5 hours/week of time into the most qualified group of people that can do this right now for making India great again in the context of AI". He added that he will top up the funding to $10 million if it beats "DeepSeek R1 on all benchmarks with rigor". Dia Rekhi contributed to this story.

AI (Model) of The Tiger

India will back an industry-academia consortium to build a local language artificial intelligence (AI) foundation model and will soon issue request for proposal (RFP) to further the effort, a senior official told ET.India will back an industry-academia consortium to build a local language artificial intelligence (AI) foundation model and will soon issue request for proposal (RFP) to further the effort, a senior official told ET. The Centre's move, which follows the US government restricting export of AI compute infrastructure and limiting use of language models that are currently dominated by American technology firms, will place India in a select group of nations including the US and China where local companies are building prominent models. Foundation models are artificial intelligence (AI) models trained on vast repositories of data, which can be used as the base for other applications. The debate over whether India should have its own foundational model or just work on open-source models "was going on for a long time, but it's no longer an either-or debate anymore. We need to do both", said the official cited above. The government will form a technical committee which will reach out to startups and academic institutions to access the technical and financial support required for such an initiative. "It has to be a consortium, we can't put all our eggs in one basket," sources said, noting that the model based on Indian languages and datasets will (subsequently) be open sourced for the larger industry to ride on. "The action by the US kind of pushes us necessarily in that direction," the person added. The outgoing Joe Biden administration passed an executive order on January 13 which restricted nations like India from importing more than 50,000 GPUs along with imposing limitations on use of AI models. On Tuesday, the new US president Donald Trump also backed an Open AI-led industry collaboration to invest $500 billion towards AI initiatives in the US. "As AI becomes more powerful, countries are likely to impose restrictions on sharing these technologies, just as we've seen with AI chips. India cannot afford to be left behind," said Srikanth Velamakanni, co-founder of Fractal.ai. He said the real reason India must build its own foundational models is to advance AGI or Artificial General Intelligence, which aims to create machines with human-like intelligence. The Centre's move comes amidst calls by Indian-origin technology leaders such as Aravind Srinivas, co-founder of one of the most valuable AI startups Perplexity for India to not "fall into the trap" of thinking that (foundational) models need a lot of money to train. The cost of training and housing of such models is steep, running into millions of dollars while also requiring high-powered compute or GPUs that are an expensive and scarce resource currently. However, the recent launch of open-source AI reasoning model by a Chinese AI startup -- DeepSeek -- has raised questions on that theory. The Chinese free, and open-source model is arguably rated as better than the ones built by American majors Meta and OpenAI and was developed at a significantly lower cost. "I hope India changes its stance from wanting to reuse models from open source and instead trying to build muscle to train their models that are not just good for Indic languages but are globally competitive on all benchmarks," Srinivas wrote on X on Tuesday. He followed it up on Wednesday with an offer to "invest $1 million personally and 5 hours/week of time into the most qualified group of people that can do this right now for making India great again in the context of AI". He added that he will top up the funding to $10 million if it beats "DeepSeek R1 on all benchmarks with rigor".