Yotta to deploy 20,736 Nvidia Blackwell Ultra GPUs in over $2 billion supercluster

Signs four-year, $1 billion commercial deal with Nvidia to host DGX Cloud cluster.

Yotta Data Services will invest over $2 billion to deploy 20,736 liquid-cooled Nvidia Blackwell Ultra GPUs and build one of Asia’s largest AI superclusters.

The infrastructure is expected to go live by August 2026.

The supercluster will be deployed at Yotta’s 60 MW D2 hyperscale data centre at its Greater Noida campus, which is scalable to 250 MW.

Additional capacity will be supported by Yotta’s Navi Mumbai campus, scalable to 2 GW.

Yotta said its platform can scale beyond one million GPUs within the next three to five years as India’s AI ecosystem accelerates.

Separately, Yotta has signed a four-year commercial engagement worth over $1 billion with Nvidia.

Under this, Nvidia will establish one of APAC’s largest Nvidia DGX Cloud cluster within Yotta’s HGX B300 Blackwell Ultra supercluster.

Nvidia DGX Cloud has been utilising Yotta’s GPU infrastructure over the past year, and the expanded deployment extends that relationship.

Over 10,000 GPUs for IndiaAI Mission

Alongside the DGX Cloud deployment, Yotta is committing over 10,000 Nvidia B300 GPUs from the supercluster to the IndiaAI Mission.

The allocation will support sovereign Indian foundation model development, research institutions, startups and population-scale public AI platforms.

The combined capital commitments, including over $2 billion in Blackwell Ultra infrastructure deployment and over $1 billion in contracted DGX Cloud engagement, reflect sustained demand for high-performance AI infrastructure in the region, the company said.

Yotta currently operates over 10,000 Nvidia GPUs in production, with another 8,000 GPUs expected to go live within the next quarter.

The company plans to deploy the 20,736 Blackwell Ultra GPUs by August 2026 and scale beyond 80,000 Nvidia GPUs by FY27 through phased infrastructure expansion and long-term capacity planning.

Infrastructure stack for advanced AI workloads

The supercluster is built on Nvidia reference architecture and integrates 800 Gbps Nvidia Quantum-X800 InfiniBand networking, advanced liquid-cooling systems, and over 40 petabytes of high-performance parallel file-system storage.

The platform is designed to support trillion-parameter foundation model training and high-throughput inference workloads capable of handling multi-million simultaneous prompts.

Yotta’s AI Factories are based on Nvidia reference architecture to deliver faster time to market, lower cost per token and higher returns.

The company said the infrastructure is engineered to support frontier-scale training and mission-critical inference deployments for global AI model developers, enterprises and governments.

Beyond infrastructure, Yotta is augmenting its Shakti Studio AI platform with Nvidia Nemotron open models, NIM microservices and access to the full AI Enterprise software suite.

Through Shakti Studio, developers in India will have access to the Nvidia Nemotron family of open models, including model weights, training datasets and recipes to enable fine-tuning and customisation.

The platform supports sovereign AI development and production-grade AI applications.