India should not chase costly LLM race; instead, it should focus on practical AI models: Mohandas Pai

India’s current constraints in capital and computing infrastructure make practical AI adoption the more sensible priority over opting to chase the expensive race to build frontier large language models (LLMs), recommended Mohandas Pai, chairman of venture capital firm Aarin Capital and former CFO at Infosys.

He was speaking at the launch of fintech firm ToneTag’s voice-first merchant banking platform eKosha here on Wednesday.

Mr. Pai further said India should focus on deploying AI to solve real-world problems along with improving productivity. “India’s AI success would ultimately be measured not by the number of frontier models it builds, but by how effectively AI improves the lives and productivity of ordinary citizens, entrepreneurs and small businesses,’’ he said.

According to him, the public discourse lacks recognition of enormous investments and infrastructure efforts required for building LLMs and the country has become overly focused on whether it was building LLMs.

“We see so many criticisms in this country by people who suddenly become experts on AI and technology, saying, why are you not building LLMs, or why are you not doing this? Building LLMs costs $35-40 billion, and nobody is writing that cheque in this country,” Mr. Pai stated.

He also said that access to massive hyperscale computing infrastructure was essential for developing the frontier AI models. To create AI, the country required hyper-cloud, and it has two gigawatts of capacity, while the US has 40 GW, and the US would be investing some $3 trillion over the next two years. “In India, we are still to start the voyage,” he commented.

Voice AI will be the future

Voice would be the next major interface for digital services, replacing many app-based interactions by enabling people to communicate with technology in the most natural way, opined Mr. Pai, adding, “Voice is the primary means by which human beings communicate.’’

The practicality of voice interfaces is exponentially increasing across Indian languages due to advances in speech recognition, speech-to-speech translation and small language models, he observed.

AI and robotics were likely to automate a significant share of human activities in the coming years, and it was essential to equip India’s more than 60 million MSMEs with affordable technologies that improve productivity, he recommended.