Free tool · no sign-up
Compute Price Converter
One megawatt of AI capacity is priced three ways: per GPU-hour on demand, per MW-year under contract, and per kW-month as a lease. Convert between them and see how the landlord's layer compares with the compute layer.
1 · GPU GENERATION
700 GPUs per critical-IT MW (author estimate; actual densities vary with configuration, networking and cooling design).
Price from the Public Rate Card Composite, H100, print 14 Oct 2026.
OR START FROM A REFERENCE PRINT
Public Rate Card Composite · on-demand list prices, all provider classes · print 14 Oct 2026
Our composite of providers' published on-demand list prices. Method: Compute Terminal Public Rate Card Composite
GetDeploying weekly median · advertised on-demand prices across providers · week of 12 Oct 2026
Advertised on-demand medians, not transaction indices. CC BY 4.0. Source: GetDeploying GPU price dataset (getdeploying.com/dataset/gpu-prices), CC BY 4.0
Lease presets appear once the Lease Comps register has approved rows; enter a $/kW-month figure meanwhile.
Enter a lease rate in $/kW-month to see how the compute layer compares with the landlord's layer on the same kilowatt.
Reverse check: at $20–25m per MW-year — the range the AI Infrastructure Primer 2026 uses for 2026 multi-year contracts — H100 at 100% billed utilisation equals $3.26–$4.08 per GPU-hour.
THE WORKING, AT THESE INPUTS
- Contract $m/MW-yr = $5.3275 × 700 GPUs/MW × 8,760 h × 100% ÷ 10^6 = $32.7m
- Per kW-month = $m/MW-yr × 10^6 ÷ 1,000 ÷ 12 = $2,722
- Multiple = compute $/kW-month ÷ lease $/kW-month = —
- Reverse $/GPU-hr = $20–25m × 10^6 ÷ (700 × 8,760 × 100%) = $3.26–$4.08
How the conversion works
Contract $m per MW-year = price × GPUs per MW × 8,760 hours × utilisation ÷ 1,000,000. Per kW-month = that figure × 1,000,000 ÷ 1,000 ÷ 12. GPUs per critical-IT MW are author estimates from rack power: H100 ~700, B200 ~520, B300 ~520, GB300 ~550, Rubin ~330. Lease presets, once available, come from the Lease Comps register and use the midpoint of each market's range. The lease layer prices power-ready space only, not the GPUs, so the two layers are not substitutes. Every calculation runs in your browser; nothing you enter is sent anywhere. Illustrative arithmetic, not a valuation and not investment advice.
Subscribers get the full three-layer stack daily by SKU and term in the Compute Price & Basis Heatmap, and the conversion assumptions are set out in the free AI Infrastructure Primer.