AI Capex & Capital Flow Workbench

Stress-test hyperscaler commitments, utility margins, and enterprise ROI payback horizons

Annualized MW OpCost $2.18M/MW Direct power + facility amort.
Net Payback Horizon 3.4 Years Capex net cash breakeven
Margin Split (Chips vs Power) 78% / 22% Hardware supplier capture
Implied Enterprise ARR $22.1B Token sales threshold

Capital Flow Allocation Conduit

Hover nodes to trace multi-year cash waterfall

Payback Horizon Sensitivity Matrix (Years to Net Breakeven)

Compute Utilization % vs Power Cost $/MWh
Power Cost \ Utilization 50% 65% 80% (Base) 90%

Capital flow arithmetic and a power-cost unit audit

Read the explanation

The workbench assigns fifty five percent of seventy five billion dollars to chips, forty one point two five billion, and twenty five percent to facilities, eighteen point seven five billion. Bars encode those assumed allocations at six pixels per billion. A separate debt node contributes seventy five times point six five times point four, or nineteen point five billion. The diagram is an authored flow illustration and does not enforce a complete balance sheet. Its enterprise flow uses capex times token price times point two eight, while the separate ARR metric uses leverage and retention as well. Missing D3 stops the original Sankey and sensitivity rendering, but sliders and export listeners were registered first and metrics update before the failure. The local check preserves the original blank visualization and actual CSV behavior. The payback formula divides leveraged capex by annual hardware depreciation times token yield times one point eight three. Substituting depreciation as leveraged capex times hardware share over lifetime cancels leveraged capex completely. Defaults give four divided by point five five times one point two five times point eight two times one point eight three, about three point eight eight, shown three point nine years. Changing capex or leverage cannot change that formula result. The separate sensitivity equation at eighty percent utilization gives about four point nine years at default power cost, so it is a different model, not the same base payback. Bars compare these rounded captions at fifty pixels per year. The margin split seventy six versus twenty four is a normalized spend share, not measured supplier profitability. The power-cost calculation exposes a unit discrepancy. One megawatt operating for eight thousand seven hundred sixty hours at sixty eight dollars per megawatt-hour costs five hundred ninety five thousand six hundred eighty dollars. That is point five nine five six eight million dollars. The code divides by one thousand and adds one point five eight, displaying five hundred ninety seven point two six million per megawatt instead of about two point one seven five six eight million under the stated units. The upper bar uses one half pixel per source-caption unit; the lower uses one hundred pixels per correctly converted million, with differing scales explicitly labeled. The calculated facility amortization variable is unused. Actual capex edits and CSV downloads preserve the source results, including this inconsistency, without correcting shared production or treating assumed ARR as observed revenue.

Super generates helpful tools and automates fact-checking across the internet proactively. If you enjoyed this tool, build your own with Super and share it with a friend.