Wholesale Compute

The operating model

Adapted from the Zaba planning model. Each assembly represents eight GPUs and 70,080 available GPU-hours per year before utilization. Hardware is modeled at $330,000 per assembly, including a 10% procurement allowance.

Uses 80% service capacity and 20% standby, with 48 service months after the selected planning delay. Capacity is an adjustable scenario input. Managed sales fees and selling costs are excluded. SLA insurance and later service funding remain to be finalized.

Costs and timing

The base budget uses $624,317.35 annual shared costs plus $26,565.79 per assembly, and monthly hosting of $53,757.81, $60,346.35 or $66,934.90 for one, two or three nine-assembly racks. A $100,000 annual certification supplement ends after 36 operating months. Figures are provisional planning allowances.

The reserve equals six months of the initial operating budget. Warranty upgrades, future hardware replacement and unpriced costs are excluded. Default prices of $5.25, $8 and $12 per GPU-hour are staged assumptions, not forecasts or current quotes.

Reading the result

Annualized gross-receipts equivalent: —. This is a mathematical comparison, before selling costs and management fees; it is not net return or cash-flow IRR.

Months run from funding. Receipts include recovery of original capital. Annualized receipts = (total receipts / initial funding)^(12 / elapsed months) − 1; this is not cash-flow IRR.

The wholesale purchase model does not deduct the operator’s costs from gross buyer collections.