'India Needs Purpose-Driven AI, Not One-Size-Fits-All:' Bharat GPT Founder on Sovereign AI

For Anksuh Sabharwal, creating a sovereign AI model in India is both about launching a product locally and about starting from India with Indian data. “India’s first sovereign AI model, what makes anything sovereign is it should be started from India, and all the ingredients should be from India. And for any model, the ingredients are the data.”

He explained that Bharat GPT builds on proprietary Indian data collected over the years, including text, voice, and video in multiple languages and dialects. “We had more than a billion users using the AI assistant created by us. We were rightly placed with the right ingredients to create the model.”

In a country with over 22 official languages and hundreds of dialects, the scale and diversity of the dataset are crucial to ensure AI understands local nuances.

Purpose Over Scale

Contrary to the global AI trend of trillion-parameter models, Sabharwal emphasized a smaller, focused approach: “Model is not the mega model of a trillion parameters. It is a small model; the smallest model we have is half a billion parameters, we build purpose. Let small models.”

He cited multiple benefits of a smaller, targeted model:

• Less energy and computing required

• Lower risk of hallucinations, since it only processes relevant, purpose-specific data

• Faster response times

• Reduced costs

“See, what we believe is the purpose is important; it also costs less. And yeah, a lot of benefits with this.” This mirrors a growing global debate; in some contexts, purpose-driven AI may be more effective than generic, massive models, especially when dealing with sensitive or localized tasks.

Rejecting the One-Size-Fits-All Model

India’s linguistic, cultural, and economic diversity makes universal AI models impractical. Sabharwal was critical of global AI attempts that try to cover all domains: “We never aspire, we never attempt to have one size fits all. Not possible. We are all so diverse. Even for India, the health sector is so diverse. One model which takes care of health and finance, and travel and legal, is not possible at all.”

He stressed that collaboration with domain experts is essential: “If we launch something with insurance, we work with the top insurance company in India, for banking, we work with financial institutions, we begin with the end in mind, we customize it.”

How Indian Models Differ from Global Giants

Comparing Bharat GPT with international models like ChatGPT or Gemini, Sabharwal stressed local context, security, and enterprise relevance, “Other models are very big, no purpose, our focus is, we want enterprises to win with our model.”

He added, “Our model now works on edge, on device, on your desk. So that much grounded. That much secure. So then there is no comparison.” Edge deployment ensures data stays on-device, addressing privacy and regulatory concerns, which is particularly relevant for government, defense, and financial institutions.

The Challenges Ahead

While smaller, purpose-built models have advantages, there are challenges in scaling adoption beyond niche enterprise applications:

• Integration with broader ecosystems may be difficult if models are highly specialized

• User expectations shaped by global models might demand capabilities beyond the purpose-driven scope

• Maintaining linguistic and cultural coverage at scale remains complex

Also read: OpenAI to Access Up to 1 GW AI Compute Capacity Through Tata Group