As the global race to develop next-generation AI models heats up, a question looms large for India: can it build its own revolutionary AI platform?
This question was the overarching theme of the panel discussion ‘Can India Build Its Own ChatGPT or DeepSeek? The GenAI Race Is On’ at the fourth edition of the ET Soonicorns Summit in Bengaluru on August 22. The panellists were Sandeep Alur, CTO – Microsoft Innovation Hub, Microsoft India; Shalini Kapoor, Strategic Advisor- Data undefined Dr Vikram Sampath, Co- Founder, NAAV AI; Ankush Sabharwal, Founder, CEO, and CTO, CoRover AI; and Debayan Dasgupta, Co-Founder, Theranautilus.
AI as a generalist
Sandeep Alur opened the conversation by highlighting a key gap: today’s AI models are “generalists” that can see, hear, speak, and reason, but they lack alignment with specific human values and cultural nuances, which are especially relevant in a linguistically and culturally diverse country like India. “The challenge is human alignment,” he noted. AI needs to reflect the personas and lived experiences of its users, not just spit out generic responses.
Ankush Sabharwal echoed this by reframing the question. “It’s not just about building the tech for the sake of it. It’s about asking ‘what’s the purpose?’ If existing models fall short by failing to address India’s unique needs in healthcare, education, or regional languages, the imperative to build locally becomes clear.”
Meanwhile, Vikram Sampath contextualised India’s linguistic diversity as an opportunity to build for the world. Unlike models trained predominantly on English or Chinese data, Indian AI can leapfrog by mastering multilingual and multicultural contexts, creating versatile yet accurate tools. His company, NAAV AI, focuses on language translation and content creation sensitive to local cultures.
“India has 22 official languages and thousands of dialects. We are uniquely placed not just for scaling faster, more accurate multilinguistic AI models, but creating them for the world,” he said.
“Indians code-switch naturally. Our models must mirror that reality, not Western or Chinese paradigms.”
India’s missing puzzle piece: A research ecosystem
Both Sampath and Shalini Kapoor brought up a structural hurdle: India’s AI research ecosystem is underdeveloped compared to the US and China. While countries like the US boast collaborative hubs, India lacks consolidated centres where well-funded teams can innovate freely on foundational AI problems.
Funding woes add to this challenge. Despite India’s massive data generation, investment in AI research has historically been low: less than 1% of our GDP goes into research overall, mostly outside tech. Encouragingly, government initiatives like the National Research Foundation have started pumping resources into cross-disciplinary AI projects. Kapoor shared how EkStep partners with premier Indian institutes, prioritising research that’s tied to India’s pressing challenges.
“India’s strength is its people. We are data consumers, yes, but data emitters too. We must create datasets that are relevant to us,” she said.
This localisation has global potential beyond India. But achieving this requires more funding, government support, and a mission-mode approach.
Infrastructure: The backbone of India’s AI ambitions
On the infrastructure front, Sandeep Alur stressed the need to scale local computing power. India’s expanding data centre market is promising, but building and deploying AI models requires vast, ethical datasets and dedicated computing resources.
Ankush Sabharwal remained optimistic: “Infrastructure and money are not insurmountable barriers. If you solve real problems with conviction, partners will come on board,” he said. He advocated for small, building mini-models focused on specific issues, and scaling up from there. This aligned with Alur’s emphasis on small language models that are precise, economical, and best suited to India’s diverse needs.
Monetisation: The real test
Building a model is only half the battle. Making it viable in India’s cost-sensitive market is another. Debayan Dasgupta highlighted the challenge of monetising AI in a space flooded with free offerings by global giants. The key lies in deep integration with existing workflows, protecting data sovereignty, and tailoring solutions to industry-specific pain points.
For example, local companies want AI that understands India’s regulatory and subsidy landscape, not generic responses from global models. By embedding AI into these unique workflows, Indian startups can carve out sustainable business models.
As the discussion wrapped up, panellists agreed that India’s greatest strength is its unique use cases and the massive data it generates. Building sovereign models trained on India-specific datasets, especially in local languages, could unleash social and economic transformation. But to do so, India must create an ecosystem that retains talent, funds innovation, and bridges the gap from lab research to scalable products. It’s a long road, but the foundation is being laid.
360 ONE is the Presenting Partner of the ET Soonicorns Summit 2025, with Shiv Nadar University as the Ecosystem Partner, Raymond as the Wardrobe Partner, Pi42 as the Gold Partner, Bank of India as the Banking Partner, Tracxn as the Knowledge Partner, and K-Tech Startup Karnataka as the State Partner. The Gifting Partners of the Summit are The Mind & Company, Plum, Clinikally, EM5, and True Elements.