Sarvam AI can train competitive models for a fraction of Big Tech’s cost, co-founder Pratyush Kumar says

When Pratyush Kumar, co-founder and chief executive of Sarvam AI, took the stage at the India AI Impact Summit in New Delhi this week, he wasn’t just unveiling products. He was making a broader argument that the notion that only a few global tech giants can build powerful artificial intelligence (AI) is already breaking down.

In an interview with Business Today, Kumar said his Bengaluru-based startup has shown that advanced models can be developed with far fewer resources than previously assumed.

“What we have managed to do is significantly lower amount of resources, train significantly useful models,” he said, adding that improvements in technology have also reduced the cost of building such systems. “We are in a deflationary world of AI itself.” 

Sarvam AI, backed by Peak XV Partners, Lightspeed Venture Partners and Khosla Ventures, has spent the past two weeks unveiling a broad suite of products, from speech and vision systems to developer tools, state partnerships and consumer hardware, in a coordinated push in time for the India AI Impact Summit 2026 in New Delhi.

The company also introduced new large language models (LLMs) to compete globally while focusing on Indian languages and data. 

Kumar argued that fears of AI becoming a monopoly controlled by a handful of companies are overstated.

“Those moats around, ‘only a couple of companies will build LLMs for the whole world,’ they are not there,” he said. 

Instead, he framed the challenge for India not as catching up to the United States or China but as ensuring the technology reaches ordinary citizens at scale.

“The goal is that the technology should be accessible to all,” Kumar said. “The diffusion from ‘I trained a model’ to ‘I’m using the model at a population scale’ that is a transition we should make.” 

To achieve that, Sarvam is betting not only on cloud-based systems but also on devices and infrastructure that bring AI directly into daily life. The startup recently showcased smart glasses designed and manufactured in India.

Kumar said such devices could give people “access to the technology, the intelligence, anywhere and everywhere,” particularly when they are on the move. 

Beyond hardware, Sarvam has been expanding across the AI stack. It launched a vision-language model for document understanding trained on multilingual data, announced partnerships with state governments to build AI-optimised data centres for public-sector useand released new open-source foundation models with large context windows aimed at developers and enterprises.

Kumar said controlling training data is central to building trustworthy systems tailored to local needs.

“A large language model is a product of the data,” he told Business Today. “Whatever it knows, the inclinations it has, the biases it has, is a function of what is in the data.” 

Despite the focus on Indian datasets, he stressed the models are not isolated from global knowledge.

“These are general-purpose models… there is a phase when you show it the world’s knowledge, and there’s a phase where you align the model to the values you consider important,” Kumar said in the interview. 

Kumar also highlighted voice as a key interface for India, where typing in many languages remains difficult and digital adoption has often leapfrogged directly to mobile and speech.

“India is a voice-first nation,” he said, noting that AI systems must work across multiple languages and noisy real-world environments. 

For enterprises, he expects adoption to accelerate as the technology becomes more reliable, pointing to rapid growth in automated voice interactions such as customer service calls.

“Voice AI will become an infrastructure,” he said. 

While Sarvam’s rapid expansion will require significant capital, Kumar said that the company has been cautious about how it deploys funding. 

“This is an area which is capital intensive,” he said. “We have been very mindful about what bets we make and how we spend capital.” 

Kumar framed the company’s mission less as building a single breakthrough product and more as establishing India as a serious AI producer.

“India should be building its own thing,” he said, arguing the world has entered a “post-AI world” where the technology will shape every sector.