☁️

GPU Cloud Rental vs Buy Calculator

Compare hourly cloud GPU rental against buying local hardware for your AI workload. Find your break-even month and the cheaper path.

Pick a cloud instance

Pick a local GPU

Workload & costs

Cloud

Storage / IP retention.

Local

PSU, case, CPU, RAM.

After projection period.

Typical: 40-90%.

Cloud total (24 mo)

Local total

Local electricity

Effective $/hour (cloud)

Break-even

Verdict

Month-by-month cost

Month Cloud cumulative Local cumulative Difference Status

Cloud instance snapshot

Instance GPU VRAM $/hr 30-day 4hr/day
RunPod RTX 4090 (Secure Cloud) RTX 4090 24 GB $1.49 $179
Vast.ai RTX 4090 (spot-ish market) RTX 4090 24 GB $0.79 $95
Lambda Cloud 1x A100 SXM A100 80GB SXM 80 GB $1.99 $239
Lambda Cloud 1x H100 SXM H100 80GB SXM 80 GB $2.49 $299
CoreWeave 1x A100 A100 80GB 80 GB $2.25 $270
CoreWeave 1x H100 H100 80GB 80 GB $2.99 $359
AWS g5.2xlarge (1x A10G) A10G 24GB 24 GB $1.01 $121
AWS p4d.24xlarge (8x A100) 8x A100 40GB 320 GB $32.77 $3932
AWS p5.48xlarge (8x H100) 8x H100 80GB 640 GB $98.32 $11798
Google Cloud A3 VM (8x H100) 8x H100 80GB 640 GB $93.24 $11189
Azure NC24ads A100 v4 A100 80GB 80 GB $3.60 $432
Azure ND96isr H100 v5 8x H100 80GB 640 GB $99.12 $11894

When cloud wins

  • • Sporadic workloads (under ~2 hours/day).
  • • Need instant access to H100/A100 without upfront capital.
  • • Teams in different time zones sharing hardware.
  • • Scaling training jobs horizontally for short bursts.
  • • Want to avoid maintenance, noise, and power constraints.

When buying wins

  • • Daily inference or fine-tuning over many months.
  • • Low electricity rates and space for a local rig.
  • • Sensitive data that must stay on-premise.
  • • Want to resell GPU later to recover part of the cost.
  • • Can use the machine for other tasks when not training.

🚀 Get AI automation insights daily

15:00 MST. One-click unsubscribe.

Subscribe