The tech industry on Thursday welcomed, electronics and IT minister Ashwini Vaishnaw’s announcement about the government’s plans to build its own foundational model on the lines of ChatGPT, DeepSeek R1, stating that the ideal way forward is to build a dual approach which focuses on both large language models as well as small language models. A dual approach is the best way to enhance the country’s AI capabilities, was the unanimous view of industry executives.
Ganesh Gopalan, co-founder & CEO of Gnani.ai, said, “India should consider focusing on both foundational LLMs and SLMs for a balanced approach”. According to him, while LLMs offer global relevance, SLMs could provide immediate benefits for localised applications, especially in sectors such as BFSI, healthcare, and customer service.
Echoing the view, Mihir Shukla CEO of Automation Anywhere, said, “I think at large there are two markets. One is the AI infrastructure market and I put Nvidia and LLM models in it which is important. But compared to that market, the 100x bigger market is AI applications. So India has a choice to go for both”.
Similarly, Sridhar Pinnapureddy, founder and CEO of CtrlS Datacenters, said, “Open-source AI models like DeepSeek-R1 represent a giant step forward towards sustainable AI development. By dramatically reducing costs and democratising access, these models enable developing nations to participate meaningfully in the AI revolution”.
LLMs are expansive AI models designed to generate human-like text at a global scale, while SLMs cater to specific industries or linguistic needs.
Sameer Dhanrajani, CEO of AIQRATE and 3AI, said, “The game for establishing AI supremacy is not just about LLMs, it’s all about speed and accuracy. Building a specific, cliche domain AI applications will not work. For India, this is the time to build the large language model which is India-centric and India-nuanced”.
While India has so far been slow to enter the LLM race, some domestic companies have already begun foundational AI model development. Startups such as Sarvam and Krutrim have developed LLMs tailored to India’s linguistic diversity. Sarvam’s flagship model, Sarvam-1, operates with two billion parameters, a fraction of DeepSeek’s scale.
Further, DeepSeek’s mixture-of-experts (MoE) approach has sparked discussions on whether it could be a viable model for India’s AI development.
“The (MoE) could provide an efficient path for India to develop competitive AI models by optimising resources and reducing computational costs,” said Gopalan.
On similar lines, Anuj Krishna, cofounder and president-technology and growth at MathCo said: “The Mixture-of-Experts approach is a game-changer. Imagine this: AI models that require far less computational power but still pack a punch. That’s exactly what MoE allows. It lets us scale up the size of the model or dataset without burning a hole in the pocket. This method gives us the chance to develop competitive AI models without needing to invest in massive infrastructure”.
Several factors have prevented India from developing a global-scale LLM. These include limited funding, access to high-quality training data, and insufficient computational resources.
Gopalan noted that a collaborative effort between the private sector, academia, and the government is needed to accelerate India’s AI ambitions. “By enhancing multilingual capabilities, optimising models for specific domains, and fostering collaboration, India can establish itself as a key player in AI,” he said.
One of the biggest barriers to India’s AI ambitions is the lack of domestic GPU infrastructure.
Bhaskar Majumdar, managing partner at Unicorn India Ventures, acknowledged India’s ongoing Nvidia GPU acquisition efforts but cautioned against over-reliance on foreign hardware.
“India is jumping on to AI mission with massive Nvidia GPU purchase plans. Indian solution makers are also thinking of AI as building it on top of NvidiaA cloud accelerators,” Majumdar said. However, he added that power-efficient, high-performance AI chips will emerge in the coming years, potentially disrupting Nvidia’s dominance.
