Shakti Bare Metal
Your AI shouldn’t compete for resources. With Shakti Bare Metal , it never will.
Shakti Bare Metal delivers dedicated GPU servers with NVIDIA HGX H100 (8 GPUs for ultra-fast training) and L40S nodes for efficient inference. With no oversubscription or virtualization, it ensures predictable, high-performance AI for enterprises and startups
Built with the Best
Train trillion-parameter models faster on dedicated,
uncompromised Shakti Bare Metal.
HGX H100 SXM5 is purpose-built for training trillion-parameter models, multimodal AI, and enterprise-scale generative AI applications.
L40S GPU nodes deliver ultra-low latency for conversational AI, recommendation systems, fraud detection, and personalized search engines.
Run weather forecasting, genomics, computational fluid dynamics, and seismic imaging workloads with near-linear scalability.
Enable streaming analytics, predictive maintenance, and anomaly detection by processing petabyte-scale datasets in real time.
Accelerate training and validation for self-driving cars, robotics, and edge AI systems with massive parallel computers.
Bare Metal Advantage
Peak AI Performance
with Shakti Bare Metal Precision
Dedicated Single-Tenant Infrastructure
No noisy neighbors or shared resources; complete isolation for performance-sensitive AI workloads.
Maximum GPU Performance
Direct access to NVIDIA HGX H100 (8x SXM5 with NVLink + NVSwitch) and L40S GPUs, with no virtualisation overhead.
Ultra-Fast Interconnects
900 GB/s GPU-to-GPU bandwidth via NVLink 4.0/NVSwitch and InfiniBand Quantum-2 (NDR, up to 32 Tbps) for seamless multi-node scaling.
Optimized for AI Training & Inference
H100 for training foundation models, LLMs, and generative AI. L40S for large-scale inference, multi-workload processing, and energy-efficient AI ops.
Scalable & Flexible Configurations
Start with a single node and scale to clusters, tailored for enterprise, research, and startup workloads.
High-Performance Storage Integration
Designed to pair with WEKA HSS parallel file system for high-throughput, low-latency data access at scale.
Supports HPC & Advanced Simulations
Optimized for emerging workloads like autonomous systems, multimodal AI, digital twins, and metaverse-scale 3D simulations.
Sovereign AI Advantage
Hosted in India, ensuring compliance, data sovereignty, and a secure AI infrastructure platform.
Peak Performance
Unveiling the Secrets of
High- Performance Architecture
- Dedicated NVIDIA HGX H100 & L40S GPUs
- Fourth-Generation NVLink 4.0 & NVSwitch Interconnects
- InfiniBand Quantum-2 NDR 400G
- Spectrum-4 400GbE Ethernet Networking
- NVIDIA BlueField-3 DPUs
- NVMe High-Speed Storage & SSD-Backed Object Storage
- Full NVIDIA AI Enterprise Stack Integration
Dedicated NVIDIA HGX H100 & L40S GPUs
Bare-metal access to the latest GPU architectures for training, inference, and visualization.
Fourth-Generation NVLink 4.0 & NVSwitch Interconnects
Up to 900 GB/s GPU-to-GPU bandwidth, ensuring faster model parallelism and reduced communication bottlenecks.
InfiniBand Quantum-2 NDR 400G
Industry-leading low latency, adaptive routing, and congestion control for large-scale AI and HPC clusters.
Spectrum-4 400GbE Ethernet Networking
Flexible high-throughput connectivity for multi-tenant or hybrid workloads.
NVIDIA BlueField-3 DPUs
Offloads networking, storage, security, and infrastructure management from CPUs, delivering faster and more secure performance.
NVMe High-Speed Storage & SSD-Backed Object Storage
Accelerated data access for training datasets, inference pipelines, and high-speed checkpoints.
Full NVIDIA AI Enterprise Stack Integration
End-to-end AI software support including CUDA, TensorRT, RAPIDS, and frameworks like PyTorch/TensorFlow.
Why Shakti Cloud Works for You
Flexible Plans for Every AI Ambition
- Monthly Price
- 6 Months
- 1 Year
- 2 Year
- 3 Year
- 4 Year
- 5 Year
|
GPU Type
|
vCPUs
|
Dedicated RAM
|
Data Drive
|
Interconnect
|
Networking |
*Monthly Price
(Per GPU Per Hour) |
|---|---|---|---|---|---|---|
|
8 x HGX H100 (640 GB GPU Memory)
|
224 | 2 TB | 61.44 TB | NVLink, NVSwitch | InfiniBand | ₹ 351 |
| 4 x L40S (192 GB GPU Memory) | 128 | 1 TB | 7.68 TB | PCIe | Ethernet | ₹ 197 |
| 8 x HGX B200 | 224 | 2 TB | 30.4 TB | NVLink, NVSwitch | InfiniBand | ₹ 473 |
| * Unit Rate: Per GPU, per Hour Please note: Data transfer (ingress and egress) charges are included in the listed price.
The cost also includes the NVIDIA BlueField-3 DPU. Shakti Cloud InfiniBand Interconnect is charged separately. Refer to AI Support Services page for more details |
||||||
|
GPU Type
|
vCPUs
|
Dedicated RAM
|
Data Drive
|
Interconnect
|
Networking |
*6 Month Price
(Per GPU Per Hour) |
|---|---|---|---|---|---|---|
|
8 x HGX H100 (640 GB GPU Memory)
|
224 | 2 TB | 61.44 TB | NVLink, NVSwitch | InfiniBand | ₹ 338 |
| 4 x L40S (192 GB GPU Memory) | 128 | 1 TB | 7.68 TB | PCIe | Ethernet | ₹ 175 |
| 8 x HGX B200 | 224 | 2 TB | 30.4 TB | NVLink, NVSwitch | InfiniBand | ₹ 473 |
| * Unit Rate: Per GPU, per Hour Please note: Data transfer (ingress and egress) charges are included in the listed price.
The cost also includes the NVIDIA BlueField-3 DPU. Shakti Cloud InfiniBand Interconnect is charged separately. Refer to AI Support Services page for more details |
||||||
|
GPU Type
|
vCPUs
|
Dedicated RAM
|
Data Drive
|
Interconnect
|
Networking |
*1 Year Price
(Per GPU Per Hour) |
|---|---|---|---|---|---|---|
|
8 x HGX H100 (640 GB GPU Memory)
|
224 | 2 TB | 61.44 TB | NVLink, NVSwitch | InfiniBand | ₹ 263 |
| 4 x L40S (192 GB GPU Memory) | 128 | 1 TB | 7.68 TB | PCIe | Ethernet | ₹ 159 |
| 8 x HGX B200 | 224 | 2 TB | 30.4 TB | NVLink, NVSwitch | InfiniBand | ₹ 473 |
| 8 x HGX B300 | 256 | 2 TB | 30.72 TB | NVLink, NVSwitch | InfiniBand | ₹ 1056 |
| * Unit Rate: Per GPU, per Hour Please note: Data transfer (ingress and egress) charges are included in the listed price.
The cost also includes the NVIDIA BlueField-3 DPU. Shakti Cloud InfiniBand Interconnect is charged separately. Refer to AI Support Services page for more details |
||||||
|
GPU Type
|
vCPUs
|
Dedicated RAM
|
Data Drive
|
Interconnect
|
Networking |
*2 Years Price
(Per GPU Per Hour) |
|---|---|---|---|---|---|---|
|
8 x HGX H100 (640 GB GPU Memory)
|
224 | 2 TB | 61.44 TB | NVLink, NVSwitch | InfiniBand | ₹ 253 |
| 4 x L40S (192 GB GPU Memory) | 128 | 1 TB | 7.68 TB | PCIe | Ethernet | ₹ 145 |
| 8 x HGX B200 | 224 | 2 TB | 30.4 TB | NVLink, NVSwitch | InfiniBand | ₹ 437 |
| 8 x HGX B300 | 256 | 2 TB | 30.72 TB | NVLink, NVSwitch | InfiniBand | ₹ 864 |
| * Unit Rate: Per GPU, per Hour Please note: Data transfer (ingress and egress) charges are included in the listed price.
The cost also includes the NVIDIA BlueField-3 DPU. Shakti Cloud InfiniBand Interconnect is charged separately. Refer to AI Support Services page for more details. |
||||||
|
GPU Type
|
vCPUs
|
Dedicated RAM
|
Data Drive
|
Interconnect
|
Networking |
*3 Years Price
(Per GPU Per Hour) |
|---|---|---|---|---|---|---|
| 8 x HGX B200 | 224 | 2 TB | 30.4 TB | NVLink, NVSwitch | InfiniBand | ₹ 378 |
| 8 x HGX B300 | 256 | 2 TB | 30.72 TB | NVLink, NVSwitch | InfiniBand | ₹ 672 |
|
* Unit Rate: Per GPU, per Hour Please note: Data transfer (ingress and egress) charges are included in the listed price. The cost also includes the NVIDIA BlueField-3 DPU. Shakti Cloud InfiniBand Interconnect is charged separately. Refer to AI Support Services page for more details. |
||||||
|
GPU Type
|
vCPUs
|
Dedicated RAM
|
Data Drive
|
Interconnect
|
Networking |
*4 Years Price
(Per GPU Per Hour) |
|---|---|---|---|---|---|---|
| 8 x HGX B200 | 224 | 2 TB | 30.4 TB | NVLink, NVSwitch | InfiniBand | ₹ 316 |
| 8 x HGX B300 | 256 | 2 TB | 30.72 TB | NVLink, NVSwitch | InfiniBand | ₹ 480 |
|
* Unit Rate: Per GPU, per Hour Please note: Data transfer (ingress and egress) charges are included in the listed price. The cost also includes the NVIDIA BlueField-3 DPU Shakti Cloud InfiniBand Interconnect is charged separately. Refer to AI Support Services page for more details. |
||||||
|
GPU Type
|
vCPUs
|
Dedicated RAM
|
Data Drive
|
Interconnect
|
Networking |
*4 Years Price
(Per GPU Per Hour) |
|---|---|---|---|---|---|---|
| 8 x HGX B200 | 224 | 2 TB | 30.4 TB | NVLink, NVSwitch | InfiniBand | ₹ 279 |
| 8 x HGX B300 | 256 | 2 TB | 30.72 TB | NVLink, NVSwitch | InfiniBand | ₹ 380 |
|
* Unit Rate: Per GPU, per Hour Please note: Data transfer (ingress and egress) charges are included in the listed price. The cost also includes the NVIDIA BlueField-3 DPU Shakti Cloud InfiniBand Interconnect is charged separately. Refer to AI Support Services page for more details. |
||||||
Frequently Asked Questions
Bare Metal servers deliver uncompromised performance by providing direct access to physical hardware with zero virtualization overhead. With complete hardware isolation, root-level control, and deterministic low-latency I/O, they are engineered for mission-critical workloads that demand raw computational power and security - such as AI/ML training, high-performance computing (HPC), large-scale data analytics, and latency-sensitive financial applications. Bare Metal infrastructure also ensures strict tenant separation, making it well-suited for compliance-driven environments requiring auditability, data sovereignty, and enhanced workload isolation.
Yes, you can seamlessly scale your resources by adding additional nodes with GPUs as your workload requirements increase. If centralized storage is needed, you can integrate high-speed or object storage solutions based on the data type - whether it’s hot (frequently accessed) or cold (infrequently accessed) data.
Yes, Shakti Cloud offers a streamlined platform to help you migrate your existing workloads to our Bare Metal infrastructure with minimal downtime. To begin, you can provision S3-compatible object storage and initiate your data transfer. Once the migration is complete, the data can be moved to High-Speed Storage (HSS), after which you can launch your Bare Metal instances with the volume attached. This approach supports efficient collaboration, especially if you’re working with multiple nodes that require shared data access. Object storage can also be leveraged for archiving or storing cold data over the long term
Enterprises training foundation models or LLMs | AI-first startups needing raw performance | Research & academia running large simulations | Government & defence requiring sovereignty
Unlike standard cloud VMs that use virtualization, Bare Metal GPU nodes on Shakti Cloud provide direct access to physical NVIDIA GPUs (H100 SXM5 or L40S). This eliminates performance overhead, noisy neighbours, and throttling, ensuring maximum throughput and lowest latency for AI/ML workloads.
We offer unlimited free ingress and egress to support running your workloads at scale.
Yes. You can scale nodes up or down as required. For long-term contracts, scaling terms depend on capacity reservations, while short-term contracts allow flexible node scaling.