Yotta Infrastructure may invest $1.5 billion for more Nvidia GPUs

Hiranandani-backed Yotta to double GPU capacity, fueling sovereign AI model development in India

Yotta Infrastructure plans to double down on India’s artificial intelligence (AI) push with a fresh $1.5-billion investment to procure another 8,000 Nvidia graphics processing units (GPUs). This is on top of its ongoing deployment for the government’s India AI Mission, Chief Executive Officer (CEO) Sunil Gupta (pictured) said.

 

Yotta, backed by real estate behemoth the Hiranandani family, has almost deployed all of the first tranche of its 8,000 GPUs. These have been given to AI startups, such as Sarvam AI and Soket AI, who are trying to build sovereign large language models (LLM).

 

The second tranche of another 8,000 GPUs has been ordered already and should be put to use by December or early next year.

 

“I have given the government an in-principle commitment that we are willing to go for another 8,000 B200 Nvidia GPUs. They are pushing me to get the ordered ones by December as they have to give all those to the new operators,” Gupta told Business Standard in an exclusive interaction.

 

If the order for the third tranche is also placed, sometime in the medium term, Yotta’s total capital expenditure would be close to $3.5-4 billion over the last few years.

 

When asked if the current tariff war with the US would impact Yotta’s investment, Gupta said: “At present, there are no restrictions on India’s access to chips from companies like Nvidia. We have been able to procure what we need. However, given the recent geopolitical developments, including the Shanghai Cooperation Organisation (SCO) summit and shifting trade positions, there is always a risk. Future decisions could impact supply. While I don’t see this as an immediate risk, it is an area we continue to watch closely.”

 

Gupta added that of the first set of GPUs, Sarvam was given about 4,100 and Soket 1,536.

 

Sarvam AI was selected by the India AI Mission to build the country’s first sovereign LLM ecosystem. That is, developing an open source 120 billion parameter AI model to enhance governance and public service access through use cases like 2047: Citizen Connect and AI4Pragati.

 

Soket will develop India's first open-source 120 billion parameter foundation model. It will be optimised for the country’s linguistic diversity, targeting sectors such as defence, healthcare, and education.

 

Other than these two startups, Gnani, which will build a 14 billion parameter voice AI foundation model delivering multilingual, real-time speech processing with advanced reasoning capabilities, will get H200 SXM and H100 SXM GPUs along with InfiniBand (IB) network from E2E Networks.

 

Gupta added that the lack of those bands has been one of the main reasons why companies have not been able to give the GPUs despite having or ordering them.

 

He added, “When you are training the models, with lots of data, you want these thousands of GPUs interconnected with each other. They call it GPU on a fabric. If they work like one, the network speed disconnecting the GPU has to match the speed of the GPU, which is a special type of network. Players who procure GPUs were supposed to invest in that InfiniBand. India AI, at this stage, is more focused towards building India’s own foundational model. So, if the providers have not put in InfiniBand, even though they have invested billions of dollars in GPUs, it is of no use.”

 

India grew its GPU capacity to 34,333, from 18,417, Union Minister for Electronics and Information Technology Ashwini Vaishnaw said in May.

 

This expanded compute capacity on Cloud is expected to provide a common computational AI platform for training and inference. It is crucial for developing indigenous foundational models and AI solutions tailored to the Indian context.

 

The government is also asking everybody to bring in the right GPU configuration, along with InfiniBand if they want to be considered as bidders. Besides Yotta, other companies are Cyfuture India, Ishan Infotech, Locuz Enterprise Solutions, Netmagic IT Services, Sify Digital Services, and Vensysco Technologies.