Startups Building Indian LLMs Also Need A Global Vision: IndiaAI Mission CEO

Even as Indian startups are being roped in by the central government to build homegrown large language models (LLMs), IndiaAI Mission CEO Abhishek Singh said that startups also need to think beyond India to compete and succeed against global giants. 

Four Indian startups — SarvamAI, Soket AI Labs, Gnani.ai and Gan.ai —  have been empanelled by the central government for the project focussing on the mission’s foundational model pillar. The four startups will collaborate with the IndiaAI team to develop and deploy indigenous functional models trained on India-specific data. 

But Singh emphasised that just because there are no guarantees of success just because an Indian startup is building LLMs for India. “They will have to ultimately compete with the best in the world. It’s not that all their revenues will ultimately come from only India. They will have to have a global vision in mind when they are training models. Initial level of support may come from the government, but that will not sustain them in the long run. Ultimately, they will have to have a business model,” Singh said during the Accel AI Summit in Bengaluru on June 4. 

Singh stressed that while Indian companies have to build up the ability to compete against global giants, even global entities —  whether it’s OpenAI or Llama — will be compelled to customise their solutions for India, which is a win-win situation. “So it will be an evolving marketplace, and it will also require Indian startups to not get complacent just because they have government support,” the IndiaAI Mission CEO added. 

VC giant Accel’s founding partner Prashanth Prakash added that more and more AI-native Indian startups have come up in the past 12 months, which are largely focussing on the application layer, but there’s also enough movement in the infrastructure and foundational layer to justify Accel’s conviction on India’s AI future. 

The Padma Shri awardee added Indian companies have the opportunity to disrupt voice-based AI models, since there is a lot of scope given the breadth and depth of languages in India. “Large scale adoption for AI in public education, healthcare and other priority areas will happen through voice. So there will be a lot of such models being built from India,” he said. 

According to Prakash, Indian startups can also find some edge over international competition when it comes to frugality. “In Silicon Valley, the cost per token and inference cost is a big focus. Where is the cost growing for these startups? Frugality at a very foundational level is what I think India startups will be able to bring to the table and change the cost dynamics of the global industry,” the Accel founding partner claimed. 

Following a call for applications in January more than 506 applications have been sent in to the IndiaAI Mission. The four startups have been selected in the first phase of the project, and more and more startups will continue to be brought on board to create a competitive atmosphere needed for innovation to emerge and thrive. 

While the IndiaAI Mission is looking at promoting innovation, Singh added that regulations will be important, primarily to ensure that any AI application that is developed is compliant with India’s legal framework and the laws as they stand, whether it is the DPDP Act or other laws. “We are not inclined to have something like the EU. We are more inclined to promoting innovation, but while having enough measures with the use of the existing legal provisions to prevent any harm that may take place.”

Singh added that India’s national compute capacity for AI-based applications and development of LLMs has crossed the 34,000 graphics processing unit (GPU) mark. 

This comes a few days after the government said it was all set to procure an additional 14,000 GPUs under the IndiaAI mission, in addition to the 18,000 GPUs already procured. Singh claimed that another 10,000 to 15,000 GPUs could join the existing capacity by the end of July 2025.