Yotta Data Services is preparing to order 50,000 NVIDIA Vera Rubin GPUs in what could become India’s largest artificial intelligence infrastructure hardware deal, with the GPU procurement estimated at about $7.5 billion including networking and associated equipment. The Indian data center and cloud infrastructure company is also planning to acquire another 45,000 NVIDIA GB300 GPUs, underscoring the scale of its planned expansion in AI computing capacity.
The Vera Rubin GPUs are expected to power Yotta’s upcoming 120-megawatt D4 data center in Delhi, which is targeted to become operational between May and August 2027. GPU deliveries are expected between March and June 2027, while Yotta estimates an additional $750 million will be required for construction of the data center infrastructure.
Yotta Plans Massive NVIDIA Vera Rubin Order
Yotta co-founder, Managing Director and CEO Sunil Gupta said the company is in the process of ordering 50,000 Vera Rubin GPUs and 45,000 GB300 GPUs. The Vera Rubin portion alone is expected to cost around $7.5 billion, including networking and related equipment.
If completed as planned, the 50,000-GPU Rubin order would be the largest single GPU procurement announced in India to date. It would also significantly expand Yotta’s position in India’s rapidly developing AI infrastructure market.
Key Details Of The Yotta GPU Deal
| Particular | Details |
|---|---|
| Company | Yotta Data Services |
| Vera Rubin GPUs | 50,000 |
| GB300 GPUs | 45,000 |
| Vera Rubin procurement value | About $7.5 billion |
| Data center construction | Additional $750 million |
| New data center | D4, Delhi |
| Data center capacity | 120 MW |
| Expected GPU deliveries | March-June 2027 |
| Expected facility launch | May-August 2027 |
| Vera Rubin process technology | 3nm |
| Vera Rubin architecture | Next-generation NVIDIA platform |
The $7.5 billion figure covers the Vera Rubin chips together with networking and associated equipment, rather than representing a simple per-GPU purchase price. The additional $750 million construction requirement takes the disclosed investment associated with the Delhi project to approximately $8.25 billion before considering other potential costs.
50,000 Rubin GPUs Would Set An Indian Record
Yotta’s planned order would be considerably larger than the 9,000 Rubin GPUs recently ordered by Hyderabad-based AM Intelligence for its AI factory.
AM Intelligence’s order was described as one of Asia’s earliest Vera Rubin deployments and forms part of a planned investment of more than $8 billion in AI compute infrastructure. Yotta’s proposed 50,000-unit order would be more than five times the size of that deployment.
India’s Major Vera Rubin Commitments
| Company / Initiative | Rubin GPUs | Location |
|---|---|---|
| Yotta | 50,000 planned | Delhi |
| AM Intelligence | 9,000 ordered | Hyderabad |
| Noetra / Japan AI infrastructure | 27,500 | Japan |
Yotta’s proposed order would also be larger than the 27,500 Rubin GPU deployment announced by Noetra in Japan, which is being developed as part of a national AI infrastructure initiative supporting robotics, digital twins and physical AI applications.
Delhi Data Center To Become AI Compute Hub
The GPUs are planned for Yotta’s D4 data center in Delhi, a 120 MW facility designed specifically to support high-density AI computing workloads.
The facility is expected to go live between May and August 2027. Deliveries of the GPUs are expected to begin several months before the planned launch, allowing Yotta to install and commission the computing infrastructure.
D4 Data Center Timeline
2026
│
├── GPU procurement process
│
▼
March-June 2027
│
├── Expected Vera Rubin GPU deliveries
│
▼
May-August 2027
│
└── D4 data center targeted to go live
The 120 MW power capacity is important because large AI clusters require substantially more electricity and cooling infrastructure than conventional data center workloads.
The growth of AI computing is therefore creating demand not only for GPUs but also for high-capacity power connections, liquid cooling, high-speed networking and specialized data center designs.
Vera Rubin Targets Agentic AI
NVIDIA’s Vera Rubin platform is designed for the next generation of AI workloads, particularly agentic AI systems that can reason through multi-step tasks, use external tools and perform actions with less human intervention.
NVIDIA says Vera Rubin delivers 10 times the agent throughput of its previous-generation Grace Blackwell platform at scale. The platform includes Rubin GPUs alongside Vera CPUs, NVLink 6 switches, ConnectX-9 networking, BlueField-4 DPUs and Spectrum-6 Ethernet technology.
Yotta’s planned deployment therefore goes beyond simply adding more conventional GPU capacity. The company is positioning the infrastructure around increasingly sophisticated AI workloads.
NVIDIA Vera Rubin Compared With Blackwell
| Metric | Vera Rubin | Previous Blackwell Generation |
|---|---|---|
| Agent throughput | Up to 10x at scale | Baseline |
| Manufacturing process | 3nm | Previous generation |
| Primary focus | Agentic AI and advanced AI factories | AI training and inference |
| High-speed memory bandwidth | More than 1,500 TB/s cited for system architecture | Lower |
| Deployment model | AI factories / large clusters | AI superclusters |
Yotta’s source disclosures describe Rubin as offering 5x faster inference and 3.5x improved training compared with Blackwell, while NVIDIA’s own materials emphasize the platform’s ability to deliver substantially greater agent throughput at scale.
Yotta Is Building A Much Larger GPU Footprint
The proposed Rubin order is part of a broader strategy rather than an isolated purchase.
Yotta announced earlier in 2026 that it would deploy 20,736 NVIDIA Blackwell Ultra GPUs at its 60 MW D2 data center in Greater Noida. The company said that deployment represented an investment exceeding $2 billion and was expected to go live by August 2026.
Yotta has also outlined a roadmap to scale beyond 80,000 NVIDIA GPUs by FY27-28.
Yotta’s GPU Expansion
| Deployment / Plan | GPU Capacity |
|---|---|
| Existing production GPUs | 10,000+ |
| Additional GPUs previously planned | 8,000 |
| Blackwell Ultra D2 deployment | 20,736 |
| New Vera Rubin plan | 50,000 |
| New GB300 plan | 45,000 |
| Longer-term target | 80,000+ next-generation GPUs |
The latest procurement plan could therefore represent a substantial acceleration beyond the company’s earlier roadmap. The 50,000 Rubin GPUs alone would be equivalent to more than twice the 20,736 Blackwell Ultra GPUs announced for D2.
Why Yotta Is Investing At This Scale
Demand for AI computing is growing as companies move from experimentation to large-scale deployment of generative AI, AI agents and other compute-intensive applications.
Yotta operates as an AI cloud and data center infrastructure provider, allowing customers to access GPU computing without having to build their own large-scale AI facilities.
Its strategy also targets sovereign AI workloads, where governments and enterprises want sensitive data and AI processing to remain within national borders.
Yotta’s existing infrastructure includes facilities in Navi Mumbai and Greater Noida. The company says its AI platform supports GPU-powered Kubernetes clusters, inference platforms, AI labs and other services for enterprise and sovereign workloads.
Networking And Cooling Are Becoming Critical
The value of Yotta’s planned procurement also illustrates how AI infrastructure is becoming more than a GPU purchasing exercise.
The $7.5 billion Rubin estimate includes networking and associated equipment. Large GPU clusters require extremely high-speed interconnects because thousands of accelerators must communicate rapidly while training or serving AI models.
Cooling is another major requirement.
AI accelerators operate at much higher power densities than traditional server workloads, increasing the importance of liquid cooling and high-efficiency data center designs. Yotta’s earlier Blackwell Ultra infrastructure includes liquid-cooling systems and 800 Gbps NVIDIA Quantum-X800 InfiniBand networking.
AI Data Center Requirements
| Infrastructure Layer | Importance |
|---|---|
| GPUs | AI training and inference |
| High-speed networking | Connects large GPU clusters |
| Liquid cooling | Handles high-density compute heat |
| Power infrastructure | Supports large-scale accelerator loads |
| Storage | Feeds data to AI workloads |
| AI software stack | Enables model development and deployment |
| Data center capacity | Provides physical operating environment |
This means the economic impact of a major GPU order extends across power, cooling, networking, construction, cloud services and data center operations.
India’s AI Infrastructure Race Accelerates
Yotta’s planned order arrives as Indian companies and infrastructure providers increasingly compete to build domestic AI compute capacity.
AM Intelligence has already placed a binding order for 9,000 Vera Rubin GPUs for its Hyderabad AI factory. Yotta’s planned 50,000-unit purchase would substantially increase India’s potential Rubin capacity once the systems are delivered.
The developments align with India’s broader objective of developing domestic AI capabilities and reducing dependence on overseas compute infrastructure.
Large domestic GPU clusters can help Indian startups, enterprises, researchers and government institutions train and deploy AI systems using infrastructure located inside the country.
Supply And Capital Requirements Remain Risks
The scale of the planned procurement also creates significant execution requirements.
Yotta must secure the GPU supply, complete data center construction, establish power and cooling capacity and attract enough customers to utilize the infrastructure economically.
The company has not yet completed the proposed 50,000-GPU order, meaning the announced figure should be viewed as a planned procurement rather than an already deployed capacity.
Timing is also important. Yotta expects deliveries between March and June 2027, while the D4 facility is targeted for launch between May and August 2027. Any delays in equipment supply, construction, power availability or commissioning could affect the schedule.
The Bigger Picture
Yotta’s planned 50,000 NVIDIA Vera Rubin GPU procurement marks a major escalation in India’s AI infrastructure race. At an estimated $7.5 billion including networking and associated equipment, the planned order would be India’s largest GPU procurement announced so far and would be more than five times the size of AM Intelligence’s 9,000-GPU Rubin order. The additional $750 million construction requirement also highlights how AI infrastructure investment increasingly extends beyond chips into power, cooling, networking and specialized data centers.
The project also reflects a broader shift toward AI factories capable of supporting agentic AI and other advanced workloads. NVIDIA is ramping Vera Rubin into full production, while companies across Asia are building increasingly large domestic compute clusters. For India, Yotta’s planned D4 facility could strengthen local access to frontier AI computing, but the ultimate impact will depend on execution, customer demand and the company’s ability to operate such a capital-intensive infrastructure platform at high utilization.
Looking Ahead
Yotta expects the Vera Rubin GPU deliveries for the Delhi project between March and June 2027, with the 120 MW D4 data center targeted to go live between May and August. The company will need to coordinate GPU deployment with data center construction, power systems, cooling and high-speed networking. If the schedule is achieved, the facility could become one of India’s largest concentrations of next-generation AI compute capacity.
The planned purchase is also likely to increase competition among Indian AI infrastructure providers as demand for domestic compute grows. Yotta already has more than 10,000 NVIDIA GPUs in production and is expanding its Blackwell-based capacity, while other operators are pursuing their own Rubin deployments. The success of these projects will ultimately depend not simply on the number of GPUs installed but on whether enterprises, AI developers and sovereign customers generate enough demand to support the economics of large-scale AI factories.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
