BharatGPT & Neysa Lay Out India’s Strong Push for AI Indepen…

At the Bengaluru Tech Summit 2025, a long conversation unfolded around India’s push for sovereign AI. The session brought together Ankush Sabharwal of CoRover, which runs BharatGPT, and Karan Kirpalani of Neysa, a rising AI cloud provider empanelled under the IndiaAI Mission.

The discussion, moderated by Ravi Jain of TDK Ventures, touched on how the last three years have seen more transformation than all previous cycles of AI development combined. Jain noted that the landscape now spans giant foundation models and tiny edge-ready ones, new clouds and established incumbents, and even the early rumble of quantum technologies. 

Meanwhile, India’s need for sovereign AI has never been more pressing. 

Sabharwal explained that CoRover is advancing this mission by making life easier for citizens through enterprise AI, with its B2B2C model. The company began with conversational systems, but now offers full-stack solutions—from language models to ready-made agentic platforms for enterprises. 

He added that their tools are already serving large volumes. “50,000 developers, researchers and enterprises have used our platform and created thousands of AI agents,” he said. Most recently, CoRover announced a partnership with Intel to roll out BharatGPT Mini for edge use cases, a move Sabharwal said would further help scale this sovereignty.

What is Neysa Up To?

Kirpalani described Neysa as a new kind of AI cloud. On the surface, its job is simple: renting GPUs in the cloud so enterprises and developers can train models or deploy them at scale. But behind that simplicity, the ambition is wider. Its motto, he said, is to help customers “use their AI as quickly as possible, as often as possible, as efficiently as possible, and in as widespread a way as possible.”

Kirpalani noted that the first reason Indians should use Neysa is sovereignty—every nation wants control over its AI stack. The second is cost, especially as scaling AI use cases on global clouds like Azure, Google Cloud or AWS becomes increasingly expensive. 

He cited one Neysa customer that processes 410 billion tokens a month, an amount that the corresponding bill on a global hyperscaler would have been impossible for them to sustain. Neysa matched the performance, offered better guarantees on token latency, and delivered it at a lower cost.

Earlier during Cypher 2025, Kirpalani told AIM that through its Velocis cloud platform, the company aims to reduce India’s dependency on foreign models and data centres. “For India, investing across the stack and reducing dependency on foreign models, hardware and data centres is vital,” he said.

To support this mission, Neysa is building a massive 400 MW AI data centre with NTT, equipped to house around 25,000 GPUs. Kirpalani said the goal is to protect India from the shocks in global supply chains. Embargoes or geopolitical tensions, he noted, can slow GPU supply, but only temporarily. 

The Mix of Hardware and Software

Models from global AI firms like OpenAI and Google are increasingly supporting Indic languages, and these companies are now eyeing India as their next big market for AI. 

Sabharwal said BharatGPT supports more languages and far more slang. Global models get the broad strokes right, but often miss the layers that matter in India. He added that they do not claim superiority across the board. Instead, they focus on domains where they have practical experience. 

He shared an example: when someone asked whether the platform supported Bhojpuri, the team assumed the answer was no. During testing, however, their Hindi model performed well enough to handle Bhojpuri because it had already captured the necessary variations needed for it to work. Sabharwal said this comes from solving real problems across Indian use cases rather than relying on general-purpose training. As a result, even global platforms sometimes use CoRover’s APIs for Indian deployments.

To prevent AI adoption from slowing in India, the entire chain must move in sync: developers, model builders, clouds and data centres. 

Since data centres have long build cycles, they must be built ahead of demand. Neysa remains silicon-agnostic in theory. In practice, however, almost everything still runs on NVIDIA because of CUDA, as that parallel computing software is the real moat for the company. 

He said global companies may enter India with force, but Indian builders understand local problems better and have more room to win.

Sabharwal said India is the best place to build AI platforms because “we produce data by just living life.” The volume of usage and the hunger for solutions give builders a natural advantage. He said India adapts to new tech faster than most places. The scale of problems is huge and the willingness to try new tools is high.

Every nation sees AI as a sovereign asset, and India is no exception. Kirpalani said the entire stack, from data centres to models to applications, must be built domestically.

Sabharwal believes AI will become as ordinary as a phone or laptop. People will use it without thinking about it. He wants CoRover to hold a significant share of that shift, supporting enterprises and improving everyday life.

Kirpalani said AI will become an invisible layer in daily tasks. Giving an example of how 4G became 5G and the internet just became fast, he said that a normal India might not have noticed it, but the tech shift was huge.

“The architectural technology change that had to happen in the backend was gargantuan. I think the measure of success of a technology—and this will be the measure of success of AI—is that the average person will wind up using it on a day to day basis,” he said.

The average person will use it without knowing they are using AI. From banks to call centres to basic digital services, AI will sit inside everything. And Neo clouds like Neysa will power that shift from behind the scenes.

The post BharatGPT & Neysa Lay Out India’s Strong Push for AI Independence appeared first on Analytics India Magazine.

Generated by RSStT. The copyright belongs to the original author.

Source