Affordable access and convenience drive growing demand for Yotta’s GPUs in India

Artificial intelligence (AI) startup Sarvam AI has launched generative AI models for enterprises and developers. The Lightspeed-backed company said that the tools and models were launched in partnership with AI infrastructure and data center firm Yotta Data Services. 

“Sarvam AI used our Nvidia H100 chips to train multiple models for different use cases, which they announced at their event. They needed access to a large number of chips to train a model with billions of parameters before the event,” Yotta’s chief executive officer Sunil Gupta told TechCircle.  

We could provide the chips on short notice, enabling them to have the model up and running in time. We are extending similar support to other companies in India building large foundation models in Indian languages, he added.

To be sure, the Hiranandani Group-owned firm announced last year in December that it had placed an order for over 16,000 of Nvidia’s H100 GPUs which are suited for AI applications. The company has already deployed 4,000 chips, and 8,000 are to be deployed by November. 

Gupta added that this ease of accessibility and ‘five to six times’ lower cost has made Yotta a favourable choice among Indian AI startups, research labs, and other institutions engaged in similar projects. He further adds that sovereignty is another factor that plays a major role in making Yotta a hit among Indian players, who may be hesitant to host their critical on a foreign cloud infrastructure provider’s platform. 

 Apart from Sarvam AI, Yotta has also collaborated with healthcare and drug discovery firm Partex. Under this partnership, Partex is utilising Yotta’s Shakti Cloud platform which is based on Nvidia H100 GPU computing infrastructure. 

That said, owing to the bigger global AI market, three-fourths of Yotta’s GPUs are still utilised by global customers, who are seeking high-quality GPUs at competitive prices due to global scarcity. “Over the last few months, the Indian market has grown, with more interest in developing foundational models in Indian languages and B2B models. As the market continues to expand, I expect the balance to shift. In a year, I anticipate 75% of my GPUs will be used by Indian customers, with 25% serving global needs. It's just a matter of time,” he added.

In March, the Union Cabinet approved the ₹10,372 crore India AI Mission to establish a computing infrastructure with over 10,000 GPUs and support the development of foundational models with capacities exceeding 100 billion parameters. This month, the Ministry of Electronics and Information Technology (MeiTy) released a request for proposal (RFP) outlining the government's initial GPU capacity requirements and the timeline for deployment. Yotta sees itself as a frontrunner for empanelment as an AI service provider on Cloud.