Article Complete
AIM
Yotta Orders 8,000 NVIDIA Blackwell GPUs to Support IndiaAI Mission
Yotta Orders 8,000 NVIDIA Blackwell GPUs to Support IndiaAI Mission
GPU Gold Rush

Yotta Orders 8,000 NVIDIA Blackwell GPUs to Support IndiaAI Mission

The company is going to invest $1.5 billion for this purpose. 

Published Updated

Key takeaways

  • Yotta Data Services invests $1.5 billion to order 8,000 NVIDIA Blackwell GPUs for India's AI sector.
  • Deployment aims to meet rising demand from startups, IITs, and research institutions developing large language models.
  • Current operations include 8,000 H100 GPUs, supporting major Indian AI models like Sarvam and Soket.
  • Yotta's commitment ensures GPUs will be operational by December to January, enhancing India's AI infrastructure.
  • Sarvam AI is utilizing 4,096 GPUs for advanced model development with extensive storage capabilities.
5 min

Yotta Data Services has ordered 8,000 NVIDIA Blackwell B200 GPUs, with the systems expected to be operational by December to January, CEO Sunil Gupta said in an exclusive interview with AIM. 

The deployment will meet growing demand from India’s AI ecosystem, including startups, IITs and research institutions building large language models (LLMs). The company is going to invest $1.5 billion for this purpose. 

The new GPU expansion is part of Yotta’s continued commitment to IndiaAI. “We ordered them about two months back, and our commitment to IndiaAI is that by December and by January, these 8,190 Blackwells will be up and live, because IndiaAI is flush with demand for making more LLMs,” Gupta said.

“Majority of the large-scale AI model development in India has happened on our GPUs…and now obviously they’re (startups) putting the model for inferencing on our GPUs,” he added.

Currently, Yotta operates 8,000 H100 GPUs, and hundreds of them are live with clients, along with 1,000 L-40s. These GPUs are being used for major Indian AI models such as Sarvam and Socket, as well as by some global clients.

 Gupta said that Sarvam AI was allocated 4,096 GPUs. “Initially, we provided about 1,600, and as more GPUs came online, we delivered the balance. They are now using all 4,096 GPUs, along with petabytes of high-speed parallel file storage and object storage, supported by a full NVIDIA DGX Cloud Lepton software layer we set up for them,” he said.

On Soket AI, Gupta said the allocation happened just a few days ago. “They have been allocated 1,536 GPUs. Right now, it’s mostly for initial planning and dataset building, but in the coming days, they will start using them actively. We have kept these GPUs reserved for them.”

To simplify access for enterprises and startups that may not want to manage hardware directly, Yotta has created a platform-as-a-service layer on API. 

He added that Yotta’s remaining GPU capacity is being used by other commercial customers, both domestic and international. Looking ahead, Gupta said, “Once we get the new Blackwell GPUs approved for India, they will be deployed for these critical LLM projects.”

Membership

Stay ahead with AI insights.

Unlimited access to premium articles, exclusive analysis, and the stories shaping India's AI economy.

Subscribe now

AIM Americas

1190 Miraloma Way Suite P, Sunnyvale, CA 94085

AIM India

1st Floor, Sakti Statesman, Bellandur, Bengaluru – 560103

© 2026 AIM. All rights reserved.