Cumulative cash position
Starts at −capex; climbs by each month’s net profit as rental prices decline at your set rate. Hover for monthly detail.
Monthly economics — month 1
Gross rental revenue versus each operating cost, before rental-price decline kicks in.
US market rental prices — reference
Per GPU per hour, on-demand, as of July 15, 2026. Editable snapshot, not a live feed — click a price to load it, then verify against the sources below.
| GPU | Marketplace low | Neocloud median | Hyperscaler |
|---|
Notes: 1–3 yr reserved contracts typically run 30–50% below on-demand. GB200/GB300 NVL72 are quoted per GPU here, but vendors mix per-GPU / per-superchip / per-rack units — always confirm the unit before modeling. B300 and MI355X prices are volatile (B300 on-demand up ~136% since Nov 2025; MI355X up ~232% since Oct 2025). AMD Instinct capacity is concentrated in specialist clouds (TensorWave, Hot Aisle, Vultr, CoreWeave); Intel Gaudi 3 is effectively IBM Cloud only (~$60/hr per 8-accelerator node ⇒ $7.50/GPU·hr list), so its three tiers are shown flat.
How the math works
Revenue
GPUs × price × 730 h/mo × utilization. Price decays continuously at your annual decline rate — the dominant risk for payback, since H100/A100 rates keep falling as Blackwell supply ramps.
Operating costs
Electricity (nameplate W × PUE × 730 h × $/kWh) + facility fee on IT kW + maintenance as % of capex + platform fee on revenue + fixed monthly overhead.
Payback & ROI
Simulated month by month for 10 years. Payback is when cumulative net profit covers capex. 5-year ROI = (60-month cumulative profit − capex) / capex. No residual hardware value is assumed.