As artificial intelligence transitions from experimentation to real-world deployment, the conversation is increasingly shifting from models to strategy, infrastructure, and competitive advantage. For India’s emerging AI ecosystem, this moment is defined not only by technological progress but by a deeper structural question: whether the country will remain a consumer of globally developed AI systems or evolve into a creator of sovereign, industry-specific solutions.
In this conversation with BW Businessworld, Bharat Shankar, Co-Founder of Gnani.AI, reflects on the changing contours of the AI landscape, the rise of verticalized AI models, the persistent challenges around accuracy and trust, and why compute infrastructure may ultimately determine the pace of enterprise adoption.
What structural shift do you think will reshape the AI industry over the next few years?
I think from discussing AI behind closed doors over the years, the evolution of AI, if you have taken note of it, AI was not for everyone. That was the sort of mindset. From that to so many people now trying to build a lot of AI technology from India, that has been the major change. This is only possible not just by the mindset of MSMEs and startups, it also requires substantial support from government and industry for these companies to blossom and create something sustainable.
There was also a debate, especially in the Indian context. The debate was whether India will be a consumer of AI or creator of AI. This is a fundamental question to be addressed. Over the years, we have consumed AI that is created by the West and European countries. But now the stage is set for us to create something of our own. This is where sovereign AI comes into the picture. With availability of data and availability of compute, people are able to take this challenge and research and create AI that can be not just demo grade but deployed at scale either in enterprises or in the government sector.
Where do you see sustainable competitive advantage emerging for AI companies today?
The advantage is knowing your verticals, knowing your target areas, knowing your customers if you are building for enterprises, and building verticalized AI that performs better than generally available AI or models. Building something very specific to solve specific problems for industries is more sustainable in my opinion. That is the shift that has happened from consuming AI to creating something very specific.
What misgivings or misinformation about AI do you still see today?
That basically stems out of using very large models like GPTs. The misinformation that exists gets into the model, and when the model starts hallucinating and gives inappropriate responses, if you as a user are not able to differentiate between an actual response and a hallucinated response, you are going to be misled naturally. The way to solve this is to build grounded models where you are able to reason out the source of information. Responsible AI becomes important. Models should be able to cite where the source of information lies. They should provide accurate information and not false positives.
What are the key bottlenecks slowing AI adoption among enterprises?
There are concerns to be addressed. For example, factual correctness of AI systems is being debated. Specialized AI models are trying to solve this naturally. Then there is availability of compute. Although many GPUs are being produced, the shift has to happen at the industry level. Enterprises have traditionally had data centers running on CPUs, not GPUs. If you want to run large AI software solutions and models, you need highly capable GPU infrastructure. The shift from traditional legacy data centers to GPU-enabled AI-first data centers has to happen at a much more rapid pace for AI adoption to exponentially increase. |
