Compute

STN compute services are delivered as fully managed, private cloud infrastructure - designed for teams that need the flexibility of cloud without the tradeoffs of shared tenancy, unpredictable costs, or fragmented operations.
STN offers two primary compute capabilities:
  • GPU Compute
    Delivered as reserved, dedicated infrastructure for AI workloads
  • CPU Compute
    Delivered as flexible, consumption-based private cloud compute

GPU Compute (Reserved)

STN GPU infrastructure is offered as reserved, fully dedicated compute. Pricing is shown per GPU, per hour and varies by contract length. Customers deploying larger clusters or committing to multi-year terms can achieve materially better economics through custom pricing.
GPU Model
GPUs per Node
GPU Memory
vCPU / System RAM
Local NVMe
Interconnect
1 Yr
$/GPU/hr
2 Yr
$/GPU/hr
3 Yr
$/GPU/hr
NVIDIA B200
(HGX B200)
8
180 GB HBM3e
Tiered allocations available
31.4 TB NVMe
NVIDIA NVLink (5th Gen)
$3.75
$3.50
$3.25
NVIDIA B300
(HGX B300)
8
288 GB HBM3E
Tiered allocations available
61.4TB NVMe
NVIDIA NVLink (5th Gen)
$4.15
$3.95
$3.75
NVIDIA GB300
(GB300 NVL72)
72
(NVL72)
37TB total
(GPU HBM + Grace Memory)
NVIDIA Grace (Arm)
Fabric-attached storage architecture
NVLink Switch fabric
Contact STN
Contact STN
Contact STN
  • Pricing reflects reserved, dedicated capacity
  • Larger clusters and multi-year commitments qualify for custom pricing
  • Detailed hardware specifications are available upon request
Get a Custom GPU Quote

CPU Compute (Private Cloud)

STN CPU compute delivers the same core functionality customers expect from public cloud services like AWS EC2 - deployed as fully managed, private cloud infrastructure.
Rather than offering fixed instance types, STN provides flexible, workload-shaped CPU compute. CPU, memory, and storage are tailored to application requirements and billed on a consumption-based model, giving teams the freedom to scale without being constrained by predefined SKUs.

What this enables:

  • Flexible CPU and memory configurations shaped to your workload
  • Consumption-based pricing without fixed instance types
  • Private, dedicated infrastructure
  • Seamless integration with STN GPU clusters
  • API-driven provisioning of compute resources
  • Enterprise-grade SLAs, security, and operational support
For teams familiar with AWS EC2:
  • Same operational flexibility and deployment experience
  • No shared tenancy
  • No egress fees
  • Delivered as private cloud, fully managed by STN

Reserved Clusters

Customers deploying large GPU clusters, multi-rack environments, or committing to multi-year agreements typically achieve significantly better pricing than standard per-GPU rates.
  • Large-scale training clusters
    Multi-node deployments optimized for distributed training workloads
  • Multi-node NVLink and high-bandwidth fabrics
    Advanced interconnect topologies for maximum throughput
  • Dedicated environments for regulated workloads
    HIPAA, SOC 2, and compliance-ready infrastructure
  • Long-term capacity planning
    Predictable costs and guaranteed availability for strategic initiatives
Talk to an STN Architect

Storage Services

Scalable, high-performance storage designed for AI training, inference, and data-intensive pipelines.
Storage Tier
Performance Profile
Price
Tier 1 – Performance Optimized
High-throughput, low-latency
$50 / TB / Month
Tier 2 – Capacity Optimized
Mid-performance, high-capacity NVMe NFS / S3
$25 / TB / Month

Networking & Interconnect

All networking is included at no additional cost with STN compute services. STN does not charge for data egress. Our networking is designed to support distributed AI workloads with predictable performance.
AI East-West Fabric
800 Gbps bandwidth
1:1 oversubscription
INCLUDED
AI North-South Fabric
200 Gbps bandwidth
4:1 oversubscription
INCLUDED
  • NVLink is used for intra-node GPU communication
  • Ethernet fabrics support inter-node scaling
  • No bandwidth-based billing
  • No egress fees

Managed Kubernetes

STN provides fully managed Kubernetes environments designed to run on private CPU and GPU infrastructure, without the operational overhead or hidden costs commonly associated with public cloud platforms.
Kubernetes at STN is tightly integrated with our private cloud compute, networking, and storage layers - allowing teams to run production workloads with predictable performance and cost.

Capabilities include:

  • GPU-aware scheduling and resource isolation
  • Integration with STN reserved GPU clusters and private CPU compute
  • Secure, production-grade control planes
  • Support for single-tenant or dedicated Kubernetes environments
  • Optional hybrid Kubernetes and Slurm configurations

This service is ideal for teams that want the Kubernetes experience they are used to, without becoming infrastructure operators.

Discuss Managed Kubernetes

Built as Private Cloud - Not a CSP Overlay

STN delivers CPU, GPU, storage, networking, and orchestration as a single, integrated private cloud platform.

Unlike public cloud services, STN provides:

  • Fully dedicated infrastructure
    No shared tenancy or noisy neighbors
  • Predictable pricing without egress fees
    Transparent costs, no hidden charges
  • Enterprise-grade networking included by default
    No additional networking fees
  • A single operational model across CPU and GPU workloads
    Unified management and operations
  • Direct access to STN infrastructure engineers
    Expert support when you need it

This approach allows teams to focus on building and scaling AI-driven products without managing the complexity of cloud infrastructure.

Not Sure Where to Start?

Whether you're migrating off public cloud, training your first production model, or scaling globally — our team can help design the right cluster and operational plan for your team.


Compute

STN compute services are delivered as fully managed, private cloud infrastructure - designed for teams that need the flexibility of cloud without the tradeoffs of shared tenancy, unpredictable costs, or fragmented operations.

STN offers two primary compute capabilities:
GPU Compute
Delivered as reserved, dedicated infrastructure for AI workloads
CPU Compute
Delivered as flexible, consumption-based private cloud compute

GPU Compute (Reserved)

STN GPU infrastructure is offered as reserved, fully dedicated compute. Pricing is shown per GPU, per hour and varies by contract length. Customers deploying larger clusters or committing to multi-year terms can achieve materially better economics through custom pricing.
  • Pricing reflects reserved, dedicated capacity
  • Larger clusters and multi-year commitments qualify for custom pricing
  • Detailed hardware specifications are available upon request
GPU Model
GPUs per Node
GPU Memory
vCPU / System RAM
Local NVMe
Interconnect
1 Yr
$/GPU/hr
2 Yr
$/GPU/hr
3 Yr
$/GPU/hr
NVIDIA B200
(HGX B200)
8
180 GB HBM3e
Tiered allocations available
31.4 TB NVMe
NVIDIA NVLink (5th Gen)
$3.75
$3.50
$3.25
NVIDIA B300
(HGX B300)
8
288 GB HBM3E
Tiered allocations available
61.4TB NVMe
NVIDIA NVLink (5th Gen)
$4.15
$3.95
$3.75
NVIDIA GB300
(GB300 NVL72)
72
(NVL72)
37TB total
(GPU HBM + Grace Memory)
NVIDIA Grace (Arm)
Fabric-attached storage architecture
NVLink Switch fabric
Contact STN
Contact STN
Contact STN

CPU Compute (Private Cloud)

STN CPU compute delivers the same core functionality customers expect from public cloud services like AWS EC2 - deployed as fully managed, private cloud infrastructure.
Rather than offering fixed instance types, STN provides flexible, workload-shaped CPU compute. CPU, memory, and storage are tailored to application requirements and billed on a consumption-based model, giving teams the freedom to scale without being constrained by predefined SKUs.
What this enables:
Assess

Flexible CPU and memory configurations shaped to your workload

Assess

Consumption-based pricing without fixed instance types

Assess

Private, dedicated infrastructure


Assess

Seamless integration with STN GPU clusters

Assess

API-driven provisioning of compute resources

Assess

Enterprise-grade SLAs, security, and operational support

For teams familiar with AWS EC2:

✓ Same operational flexibility and deployment experience
✓ No shared tenancy
✓ No egress fees
✓ Delivered as private cloud, fully managed by STN

Reserved Clusters

Customers deploying large GPU clusters, multi-rack environments, or committing to multi-year agreements typically achieve significantly better pricing than standard per-GPU rates.

Large-scale training clusters

Multi-node deployments optimized for distributed training workloads


Multi-node NVLink and high-bandwidth fabrics

Advanced interconnect topologies for maximum throughput

Dedicated environments for regulated workloads

HIPAA, SOC 2, and compliance-ready infrastructure

Long-term capacity planning

Predictable costs and guaranteed availability for strategic initiatives

Talk to an STN Architect

Storage Services

Scalable, high-performance storage designed for AI training, inference, and data-intensive pipelines.
  • Billed based on consumption
  • Designed for high concurrency and sustained throughput
Storage Tier
Performance Profile
Price
Tier 1 – Performance Optimized
High-throughput, low-latency
$50 / TB / Month
Tier 2 – Capacity Optimized
Mid-performance, high-capacity NVMe NFS / S3
$25 / TB / Month

Networking & Interconnect

All networking is included at no additional cost with STN compute services. STN does not charge for data egress. Our networking is designed to support distributed AI workloads with predictable performance.

  • NVLink is used for intra-node GPU communication
  • Ethernet fabrics support inter-node scaling
  • No bandwidth-based billing
  • No egress fees
AI East-West Fabric
800 Gbps bandwidth
1:1 oversubscription
INCLUDED
AI North-South Fabric
200 Gbps bandwidth
4:1 oversubscription
INCLUDED

Managed
Kubernetes

STN provides fully managed Kubernetes environments designed to run on private CPU and GPU infrastructure, without the operational overhead or hidden costs commonly associated with public cloud platforms.
Kubernetes at STN is tightly integrated with our private cloud compute, networking, and storage layers - allowing teams to run production workloads with predictable performance and cost.
Discuss Managed Kubernetes

Capabilities include:

01 - GPU-aware scheduling and resource isolation
02 - Integration with STN reserved GPU clusters and private CPU compute
03 - Secure, production-grade control planes
04 - Support for single-tenant or dedicated Kubernetes environments
05 - Optional hybrid Kubernetes and Slurm configurations

This service is ideal for teams that want the Kubernetes experience they are used to, without becoming infrastructure operators.

Built as Private Cloud
- Not a CSP Overlay

STN delivers CPU, GPU, storage, networking, and orchestration as a single, integrated private cloud platform.
Unlike public cloud services, STN provides:

Fully dedicated infrastructure
No shared tenancy or noisy neighbors
Predictable pricing without egress fees
Transparent costs, no hidden charges
Enterprise-grade networking included by default
No additional networking fees
A single operational model across CPU and GPU workloads
Unified management and operations
Direct access to STN infrastructure engineers
Expert support when you need it

Not Sure Where to Start?

Whether you're migrating off public cloud, training your first production model, or scaling globally — our team can help design the right cluster and operational plan for your team.

Quick Links
Support Services
|
Master Services Agreement
© 2026 STN, Inc. All rights reserved.
Built By