AI Capex, Decoded — Trillions Announced, Financing Engineered
AI’s trillion-dollar boom is not raw cash.
It is engineered finance.
$100B headlines move markets.
Ten gigawatts is the technical claim.
That equals the electricity load of New York City.
The capacity is not available today.
The mechanics are plain:
Equity values convert into debt.
Contracts are matched to GPU life cycles.
Public programs and sovereign pools underwrite gaps.
Vendors extend credit and residual-value options.
One real outlay appears multiple times in public accounts.
Suppliers show backlog.
Banks show pipeline.
Governments show inflow.
Yes — user inferencing and curiosity drive utilization today.
But commitment at sovereign scale is set by engineering: electrons, cooling, chips, permits, and skilled labor.
Curiosity creates pressure. Physics and balance sheets set timelines.
Why it matters:
Wall Street supplies collateral.
Washington supplies policy liquidity.
Sovereigns supply the backstop.
This is the operating reality of the build-out: financing design, not free cash flow.
If valuations compress, if subsidies thin, or if grids stall, the announced figures resolve into a different ledger.
What allocators should watch:
Bond issuance by Mag-7 suppliers.
Signed 4–5 year compute offtake contracts.
Interconnect queue approvals and transmission permits.
DOE/sovereign loan disbursements and tax-credit allocations.
Vendor financing terms and used-GPU market floors.
(These are the receipts. They separate talk from funded execution.)
Logged September 25, 2025 — held and timestamped before the cycle names its true governor: financing engineered to scale, constrained by physical science.