Cart total (0 items):
Billed monthly: per month
Sales tax (if applicable): Calculated on invoice
Key specifications
Illustrations are for reference. Service scope is defined by the plan you choose.
Pay monthly or annually — the contract value is the same.
Save 5% with a 2-year term or 10% with a 3-year term.
Dedicated 8-GPU NVIDIA H100, H200 or B200 nodes on a non-blocking InfiniBand fabric, with Slurm or Kubernetes and shared storage.
GPU Cluster Rental provides dedicated NVIDIA HGX nodes with eight GPUs each, connected by a non-blocking InfiniBand fabric, for distributed training, fine-tuning and high-throughput inference. You reserve the nodes for the full term: no spot interruptions, no shared GPUs and no per-hour metering.
Inside each node, NVLink and NVSwitch connect all eight GPUs to one another; between nodes, eight 400 Gb/s InfiniBand ports carry GPU-to-GPU traffic. Every node also has local NVMe scratch space and capacity on a shared parallel file system. We deliver the cluster with Slurm or Kubernetes, NVIDIA drivers, CUDA, NCCL and container tooling, validated by burn-in and multi-node performance tests before handover.
Our engineers monitor GPU health, replace failed components, keep firmware and drivers current in agreed maintenance windows and help your team tune multi-node jobs. You keep full administrative access to your nodes and your data.
All prices in USD, per node at the 1-year rate. Pay monthly or annually — the contract value is the same. Swipe the table sideways to see every plan.
| Feature | H100 node | H200 node Recommended | B200 node |
|---|---|---|---|
| Price | $19,000.00 /mo per node or $228,000.00/yr | $24,000.00 /mo per node or $288,000.00/yr | $34,000.00 /mo per node or $408,000.00/yr |
| With a 3-year term | $17,100.00/mo Save 10% or $205,200.00/yr | $21,600.00/mo Save 10% or $259,200.00/yr | $30,600.00/mo Save 10% or $367,200.00/yr |
| Best for | Hopper-generation nodes for large-scale training and fine-tuning. | More and faster GPU memory for long-context models and larger batches. | Blackwell-generation nodes for the largest training and inference workloads. |
| What’s included |
|
|
|
| Order |
Billed monthly · 1-year term Customize billing, term & quantity (Plan: H100 node) |
Billed monthly · 1-year term Customize billing, term & quantity (Plan: H200 node) |
Billed monthly · 1-year term Customize billing, term & quantity (Plan: B200 node) |
| SKU | TS-RNT-304 |
|---|---|
| Category | Server & GPU Rental |
| Pricing | Per node, per year; payable monthly or annually |
| Contract | Annual commitment; 1-, 2- or 3-year term |
| Billing | Monthly or annual invoicing; same contract value |
| GPUs per node | 8× H100 80 GB / 8× H200 141 GB / 8× B200 180 GB depending on tier |
| GPU interconnect | NVLink with NVSwitch: 900 GB/s per GPU (H100, H200), 1.8 TB/s per GPU (B200) |
| Cluster fabric | 8× 400 Gb/s NDR InfiniBand per node, non-blocking; redundant Ethernet front end |
| Node memory | 2 TB DDR5 (H100, H200) or 4 TB DDR5 (B200); 30.72 TB local NVMe |
| Shared storage | Parallel file system: 100 / 150 / 200 TB usable per node |
| Orchestration | Slurm or Kubernetes; root access on every node |
| Lead time | Typical delivery in 4–8 weeks, subject to hardware availability |
You can order 1 to 16 nodes (8 to 128 GPUs) per order. Nodes ordered together are placed on the same non-blocking InfiniBand fabric so multi-node jobs run at full bandwidth. Larger clusters are quoted on request.
You can choose monthly invoices, which spread the same annual commitment over 12 payments. The rental cannot be ended early: nodes are reserved for your company for the full term. You can add nodes at any time, prorated to your renewal date and subject to hardware availability.
We support the full infrastructure stack: hardware, firmware, drivers, CUDA, NCCL, the scheduler and the file system. Your model code and framework-level debugging stay with your team. The H200 tier adds a quarterly multi-node tuning session, and the B200 tier adds a named cluster engineer and a monthly performance review.
Your data stays on your nodes and file system allocation until the last day of the term; we recommend copying it off during the final two weeks. Drives are then securely wiped before the hardware is reassigned.
Have another question? Ask a solutions architect
Server & GPU Rental
Dedicated NVIDIA GPU servers on annual terms for AI training, inference, rendering and simulation, with full root access.
Up to 10% off multi-year terms
Server & GPU Rental
Fully managed virtual private servers with NVMe storage, daily backups, OS patching, hardening and 24/7 monitoring.
Up to 10% off multi-year terms
Server & GPU Rental
Single-tenant bare-metal servers with ECC memory, NVMe RAID and out-of-band access, plus hardware replacement included.
Up to 10% off multi-year terms
AI & Productivity
A governed AI coworker workspace for your company: seats, single sign-on, usage policies, onboarding and training — managed for you.
Up to 10% off multi-year terms
AI & Productivity
A dedicated GPU server running open-weight AI models for your company only: private chat, document Q&A and an internal API.
Up to 10% off multi-year terms
Software Licensing
Office 365 E3, Microsoft 365 E3 or E5 seats in 500-seat packs, billed monthly or annually, with licensing administration included.
Billed monthly or annually
Up to 10% off multi-year terms
Software Licensing
VMware vSphere Foundation or VMware Cloud Foundation core subscriptions in 512-core blocks, with license and renewal management.
Billed monthly or annually
Up to 10% off multi-year terms
Cloud, Storage & Backup
Managed, encrypted offsite backup for servers, VMs and cloud mailboxes, with immutable copies and scheduled restore tests.
Up to 10% off multi-year terms
