Davos 2026: India needs to do more on AI models, 6 to be showcased at Impact Summit, says Ashwini Vaishnaw

Minister for Electronics and Information Technology Ashwini Vaishnaw at World Economic Forum in Davos on January 22

• India to showcase 6-7 AI models at upcoming India AI Impact Summit
• Vaishnaw urges more R&D for breakthroughs in AI models
• Speech-based AI models prioritized due to India's oral culture

Did our AI summary help?

India needs to put more effort into building its own AI models, specifically in research and development, even as it prepares to showcase six to seven models at the upcoming India AI Impact Summit, Union minister Ashwini Vaishnaw said on January 22.

"I think on models we must work more,” Vaishnaw told Moneycontrol at the World Economic Forum, Davos. "The 12 models which are coming up (under the IndiaAI Mission) are a very comprehensive set of models. They are going to meet practically almost all the needs that we have."

However, he stressed that there was a need for deeper research.

"We must put a lot more effort in research and development. The next big breakthrough will come either from math or from engineering," he explained.

Vaishnaw defended India’s decision to work across multiple layers of AI development. "Given the fact that today we are on the verge of becoming the third largest economy, and given the talent that we have, it makes sense for us to work on all the five layers," he said.

On sovereign models, he said scale alone should not be a deterrent. "Today, a 50 billion parameter model does practically 95 per cent of the things that trillion parameter models would be doing," he said.

Among the models expected to be showcased, Vaishnaw pointed to Sarvam as being at an advanced stage.

"Sarvam is close to 120 million parameters. In terms of capabilities, that would be 96–97 per cent of the frontier models from the world," he said.

He also highlighted speech-based models as especially important for India. "Our culture has been very oral, very audio-based. That's why we put a lot of focus on models which are not text-based, but more hearing-based and speaking-based," he added.