'Western labs aren’t focusing on this': PerplexityAI CEO Aravind Srinivas tells Nikhil Kamath Indians can fill this major AI gap

India must develop its own AI models and compete globally in artificial intelligence, said Aravind Srinivas, CEO of Perplexity AI. Highlighting India's untapped potential in AI development, Srinivas emphasized the need for homegrown innovation rather than relying solely on foreign models.

He believes India should establish its own AI research companies, akin to DeepSeek, to train foundational AI models that can excel in reasoning, thinking, and task execution. "India should have its own DeepSeek-like company that not only competes in Indian languages but also on global benchmarks. That will inspire the next generation of engineers to build the future," he said on Nikhil Kamath's WTF podcast.

How Indian startups can enter the AI space

For Indian entrepreneurs and tech leaders, Srinivas laid out a step-by-step approach to entering the AI race: "Raise more money, get more users, and raise a little more money. Start with post-training on open-source models, then look into pre-training, and finally, invest in data centers." He emphasized that AI development is a multi-stage process, but India must start somewhere to build world-class AI systems.

Srinivas pointed out a major gap in AI development—voice recognition and synthesis for Indian languages. He noted that current AI systems struggle with Indian accents and dialects, creating an opportunity for Indian startups. "Most AIs are pretty bad at Indian voices. Speech recognition and synthesis aren't great, and western labs don’t prioritise this. But real-time AI voice synthesis that supports all Indian languages, accents, and grammar would be a game-changer," he explained.

While acknowledging the challenges in collecting and training AI voice models, Srinivas emphasized that focusing on India's unique linguistic diversity could give Indian AI companies a competitive edge.