Skip to content
6 Cost Centers · CSV Export · Shareable URL · Jul 2026

AI Infrastructure Budget Planner

A full-stack AI budget model built for CFOs and engineering leaders. Goes beyond GPU compute to cover every cost center — training, inference serving, storage tiers, egress, MLOps tooling, and fully-loaded team cost. Export to CSV for your next board deck or share with a link.

✓ Training + inference compute✓ Hot + cold storage tiers✓ MLOps tool catalog✓ CapEx vs OpEx classification✓ Shareable URL state✓ CSV export
Monthly Total — All-In
$82.8K
$993.8K/yr·$2.98M 3yr
Cost Breakdown
Training ComputeOpEx
$6.0K7%
Inference ServingOpEx
$4.8K6%
StorageOpEx
$1.1K1%
Data TransferOpEx
$8911%
MLOps ToolingOpEx
$1.3K2%
TeamCapEx-adj.
$68.8K83%
Classification
Cloud OpEx
$14.1K
/mo · compute + storage
CapEx-adj. (team)
$68.8K
/mo · fully-loaded
OpEx / CapEx split17% / 83%

All inputs encoded in URL — share with teammates or finance

01

Training Compute

$6.0K/mo
8 GPUs
1512
300 hrs/mo
0720
8 GPUs × 300 hrs × $2.49/hr = $6.0K/mo
02

Inference Serving

$4.8K/mo
4 GPUs
164
16 hrs/day
124
4 GPUs × 16 hrs × 30 days × $2.49/hr = $4.8K/mo
03

Storage

$1.1K/mo
20 TB
0500
100 TB
01,000
10 TB
0200
Hot tier — weights + artifacts
30 TB × 1,024 × $0.023/GB
$707/mo
Cold tier — datasets
100 TB × 1,024 × $0.004/GB
$410/mo
04

Data Transfer & Networking

$891/mo
10 TB/mo
0500
10 TB × 1,024 GB × $0.087/GB = $891/mocloud avg · ingress typically free
05

MLOps Tooling

$1.3K/mo
Custom Tool (optional)
$
/mo
06

Team — Annual Fully-Loaded Cost

$68.8K/mo
ML Engineers
$220K/yr fully-loaded · default 2
MLOps / Infra Engineers
$200K/yr fully-loaded · default 1
Data Engineers
$185K/yr fully-loaded · default 1
(2×$220K + 1×$200K + 1×$185K) ÷ 12 = $68.8K/mo
classified CapEx-adjacent — internal headcount investment

Methodology & Assumptions

Training & Inference Compute
  • H100 SXM: $2.49/GPU/hr — cloud avg on-demand
  • H200 SXM: $3.99/GPU/hr
  • B200 SXM: $8.00/GPU/hr
  • A100 80GB: $1.60/GPU/hr
  • AMD MI300X: $2.49/GPU/hr
  • Training = GPUs × hrs/mo × rate
  • Inference = GPUs × hrs/day × 30 × rate
Storage & Data Transfer
  • Hot storage: $0.023/GB/mo — S3-class (standard)
  • Cold/archive: $0.004/GB/mo — Glacier-class
  • Model weights + artifacts billed at hot rate
  • Training datasets billed at cold archive rate
  • Egress: $0.087/GB — cloud avg (all zones)
  • Ingress: free — not modeled
Team & MLOps Tooling
  • ML Engineers: $220K/yr fully-loaded
  • MLOps / Infra Engineers: $200K/yr
  • Data Engineers: $185K/yr
  • Divided by 12 for monthly view
  • Team classified as CapEx-adjacent
  • MLOps prices are list — negotiated rates vary