Unit Economics

Unit economics in GPU infrastructure refers to the revenue, cost, and margin analysis at the per-GPU or per-MW level. The fundamental calculation: revenue per GPU-hour (blended across spot, on-demand, and reserved pricing, net of discounts) minus operating cost per GPU-hour (power, cooling, network, maintenance, depreciation, personnel) equals margin per GPU-hour. At the facility level, revenue per MW per year minus total operating cost per MW per year determines site-level profitability. The capital efficiency metric — revenue relative to total capital deployed — determines whether the business generates adequate returns.

Key unit economics variables include: GPU hardware cost ($30-40K per H100, $40-60K per B200), useful life (36-60 months), residual value (20-40% at end of primary service), power cost ($0.04-0.12/kWh depending on geography), PUE (1.1-1.4), cooling infrastructure amortisation, network equipment cost, and personnel per MW. Revenue-side variables include pricing (on-demand vs reserved mix), utilisation, discount depth, and contract churn. A well-run neocloud targeting 50%+ gross margins needs sustained utilisation above 80% at blended pricing above cost-of-capital adjusted breakeven.

Unit economics modelling is central to our due diligence work. We build 12-scenario fleet models that stress-test revenue projections across utilisation, pricing, and contract structure assumptions. Our proprietary pricing data enables us to benchmark management assumptions against real market conditions — a capability that does not exist in any other single dataset.