Coming soon
Clusters1 cluster · 8 GPUs
8 GPUs
⌄GPU node groups
Storage tiers (TB)
Hot
Warm
Cold
Egress (TB/mo)
5 TB
☁ Cloud pricing▼
Software & licenses
Add any platform, inference server, monitoring or observability subscriptions running on this cluster.
e.g. Inference server, monitoring, orchestration…
Disaster recovery
Include DR environment
Standby cluster at secondary site
Cluster load (capacity check)
Informational only — does not change cost. A fixed cluster costs the same whether idle or busy.
75%
✓ Healthy load (75%)
⚙ Override cost rates▼
☁ Cloud other rates
Hot storage$/TB/mo
Warm storage$/TB/mo
Cold storage$/TB/mo
Egress$/TB
Control base$/mo
Supportfraction
🖥 On-prem GPU prices (CAPEX $)
H200 141GB$
H100 80GB$
A100 80GB$
A100 40GB$
L40S 48GB$
MI300X 192GB$
🖥 On-prem other costs
Depreciationyrs
Power rate$/kWh
PUE×
Colo/rack$/kW/mo
Networking CAPEX$/gpu
Staff loaded$/yr
Clusters/FTE
Support/warranty%/yr
Storage hot$/TB/mo
Storage warm$/TB/mo
Storage cold$/TB/mo
Control base$/mo
Control per GPU$/mo/gpu
✓ Healthy load (75%)
Monthly cost
☁ Cloud
$38,295
$6.65/gpu-hr eff
1.7× headline price
1.7× headline price
🖥 Self-hosted
$26,074
$4.53/gpu-hr eff
6.2× hardware price
6.2× hardware price
5-Year Cost Difference
☁ Cloud$2.30M
🖥 Self-hosted$1.56M
Difference$733K
5-year total cost
☁ Cloud$2.30M
🖥 Self-hosted$1.56M
Per-layer breakdown ☁ cloud 🖥 on-prem
GPU Compute
59%16%
Storage
24%23%
Support & Ops
11%40%
Control Plane
5%1%
Network
1%10%
☁ Cloud
1.71×
$3.89/hr headline
→ $6.65 all-in/hr
→ $6.65 all-in/hr
🖥 Self-hosted
6.20×
$0.73/hr amort
→ $4.53 all-in/hr
→ $4.53 all-in/hr
⚠ Cloud uses public list prices. On-prem uses CAPEX ÷ 5yr depreciation + power + colo + staff. Override any rate in the left panel. On-prem excludes facilities buildout and management overhead.
Export