| Sector / Node | Gross AI Capex / Debt | Circular Exposure | Collateral Haircut | Net Capital Loss | Solvency Ratio | Status |
|---|
| Sector / Node | Gross AI Capex / Debt | Circular Exposure | Collateral Haircut | Net Capital Loss | Solvency Ratio | Status |
|---|
Similar to 1999 telecom vendor financing and 2007 structured SIVs, big tech hyperscalers invest billions in frontier model developers, on the explicit condition that those cash tranches are immediately returned as prepaid cloud computing credits.
Neo-cloud operators and data center developers have secured hundreds of billions in private credit debt backed primarily by GPU clusters as physical collateral. Rapid hardware depreciation triggers mandatory collateral calls and insolvency.
Data center developers sign multi-decade Power Purchase Agreements (PPAs) with regulated utilities and grid operators. A bust in tenant monetization leaves power developers exposed with stranded generation and substation Capex.
The saved stress tester displays seven counterparties and nine links, but loss computation maps each node independently. No link amount, strength or neighbor loss enters the arithmetic. Each node adds four assumed components: circular revenue write-down, collateral loss, revenue miss and credit spread penalty. Contagion round multiplies every component by round divided by four. Under the initial McLean preset the total assigned exposure is one thousand three hundred ninety five billion, not the earlier static page label one point two five trillion. Round two gives unrounded loss three hundred seventy nine point eight two nine three seven five billion, displayed as three hundred eighty. Round four doubles the unrounded loss to seven hundred fifty nine point six five eight seven five, displayed seven hundred sixty. At point three pixels per displayed billion these bars measure one hundred fourteen and two hundred twenty eight. This is proportional scaling rather than propagated counterparty contagion. For the neo-cloud node the assumed capex is one hundred seventy five billion, circular share forty percent, equity thirty five billion and debt one hundred forty billion. At round four the circular loss is forty two billion. Collateral loss is forty nine point zero eight seven five, revenue loss thirty six point seven five and spread penalty fifteen point six eight. Their sum one hundred forty three point five one seven five exceeds the equity buffer, so the node is labelled insolvent. At two pixels per billion the loss rises two hundred eighty seven point zero three five pixels against equity seventy, a two hundred seventeen point zero three five excess. The stored node values round loss to one hundred forty four and remaining equity to minus one hundred nine. Solvency ratio displayed in JSON is clamped to zero even when the raw ratio is negative. These figures are assigned scenario inputs and outputs, not actual financial statements or investment forecasts. Inject Backstop subtracts up to fifty billion of loss from exactly three named nodes: hyperscalers, private credit and neo-clouds. At the initial round two each eligible node exceeds fifty, so total deduction is one hundred fifty billion and rounded total falls from three hundred eighty to two hundred thirty. At two pixels per billion the three equal fifty deductions measure one hundred pixels each, totaling three hundred. The collateral and circular aggregate loss readouts remain gross and do not include that deduction. Advancing beyond round four wraps back to one rather than carrying historical losses. Reset briefly assigns round one but its preset loader immediately sets round two, making actual reset round two. Export downloads the current local JSON snapshot and native SVG renders the assigned graph; no real liquidity, default, investment or regulatory action occurs.