// Interactive

Build your bundle.

Pick a workload, dial in scale — we recommend the GPU, storage, and networking mix, and show your estimated monthly spend.

// Configure
Workload
Scale — Small
PrototypeSmallMediumLargeFrontier
Priority
Commitment
Add-ons
Estimated monthly
$33,244
H100 80GB × 16 · Reserved 1-mo
− $7,578 vs on-demand baseline
Compute$26,867
Storage$3,768
Networking$609
Support$2,000
Total$33,244
Get a formal quote

// Encodes workload, scale, priority, commitment & add-ons in the URL.

// Illustrative. Actual pricing confirmed on quote.

// Recommended package

Your bundle, itemized.

Compute
16 × H100 80GB
Effective rate
$2.33 /GPU-hr
Utilization
720 hr/mo
Fabric
NVLink + InfiniBand
Monthly
$26,867
View H100 80GB
Storage
72 TB provisioned
Object (S3)
50 TB · $922
Block NVMe
2 TB · $184
Lustre FS
20 TB · $2,662
Monthly
$3,768
Storage products
Networking
2 TB egress · 1× interconnect
Egress
2 TB · $61
Interconnect
1 port · $540
Public IPv4
2 · $7
Monthly
$609
Platform overview
// Why this bundle
  • H100 80GB matches the AI Training workload at balanced priority — H100 remains the mainstream training workhorse with the best price-to-throughput.
  • 16 GPUs is our reference footprint for the Small scale of this workload.
  • Storage weighted toward parallel Lustre for training-set throughput.
  • Reserved 1-mo pricing applied — commit longer to unlock deeper discounts.
Assumptions used for this estimate
// Hours & utilization
Workload profile
AI Training
Billable hours / month
720 hr
Utilization vs 24×7
100% (of 720 hr)
GPU count
16
GPU-hours / month
11,520 GPU-hr
On-demand list rate
$2.99 /GPU-hr
Reserved 1-mo rate
$2.33 /GPU-hr (−22%)
Hours per month come from the workload profile: training and inference assume 720 hr (24×7), HPC 480 hr, rendering 300 hr, fine-tuning 240 hr. Adjust workload or scale to change.
// Savings baseline
LineOn-demandThis bundle
Compute$34,445$26,867
Storage$3,768$3,768
Networking$609$609
Support$2,000$2,000
Total$40,822$33,244
Savings vs baseline$7,578 (19%)
Baseline = the exact same footprint priced at the on-demand list rate for H100 80GB ($2.99/GPU-hr × 16 GPUs × 720 hr). Storage, networking, and support are held constant, so savings reflect only the compute commitment discount.
compute = rate × GPUs × hours
storage = Σ (TB × 1024 × $/GB-mo)
network = egress + interconnect ports × 720 hr + IPv4 × 720 hr
// Unit rates used
Object
$0.018/GB-mo
Block NVMe
$0.09/GB-mo
Lustre FS
$0.13/GB-mo
Egress
$0.03/GB
Interconnect
$0.75/port-hr
Public IPv4
$0.005/hr

// Illustrative rates. Reserved commitments assume steady utilization for the term. Actual pricing confirmed on quote.

Saved scenarios (0/8)

// Snapshots the current inputs (workload, scale, priority, commitment) plus derived hours, utilization, GPU count, storage TB, and egress TB. Stored locally in this browser.

Metric
Current
WorkloadAI Training
Prioritybalanced
CommitmentReserved 1-mo
GPUH100 80GB
GPU count16
Hours / mo720 hr
Utilization100%
GPU-hours11,520
Effective $/GPU-hr$2.33
Storage72 TB
Egress2 TB
SupportIncluded
Compute$26,867
Storage $$3,768
Network $$609
Monthly total$33,244
Savings vs on-demand− $7,578 (19%)
Sensitivity analysis

How savings shift with utilization & reserved mix

Grid below prices 16× H100 80GB at each combination. Rows step through GPU utilization (share of the 720 monthly hours). Columns step through what share of GPUs sit on the Reserved 1-mo tier vs. on-demand list. Fixed costs (storage, networking, support = $6,377) are held constant.

Utilization range (% of 720 hr)
Reserved mix range (% on Reserved 1-mo)
util ↓ / reserved →0%25%50%75%100%
40%0.0%3.8%7.5%11.3%15.0%
55%0.0%4.1%8.2%12.3%16.5%
70%0.0%4.3%8.7%13.0%17.4%
85%0.0%4.5%9.0%13.5%18.1%
100%0.0%4.6%9.3%13.9%18.6%

Darker cell = larger savings vs. an all-on-demand footprint at the same utilization. At 0% reserved every cell is 0% saved (list price on both sides). Savings scale linearly with reserved mix at the Reserved 1-mo discount (22%), and absolute $ savings scale with utilization.