The 08/2026 network diagram of the AI trade, legend doing the work: $46B of equity actually deployed (red arrows) versus $879B of multi-year purchase commitments (blue) — roughly 19 dollars of promise per dollar of cash. Microsoft→OpenAI $250B of commitments against $13B equity; Oracle↔OpenAI $300B; OpenAI→AMD $90B; Amazon→Anthropic $100B against $13B; everything routing through CoreWeave at the bottom.
The loop shape is the point: chipmaker invests in the lab, the lab commits to buy compute from the hyperscaler, the hyperscaler buys the chipmaker’s chips. One dollar of equity spawns several dollars of booked future revenue, and the same expected AI revenue supports several balance sheets at once. The rhyme is 1990s telecom vendor financing — Lucent and Nortel lending customers the money to buy their gear (memory reference) — which worked until the customers’ revenue didn’t arrive.
Why it belongs next to the credit notes: the commitments are the collateral story behind the debt wave (Big Tech becomes a borrower), and the diagram is the contagion map — if one node’s revenue disappoints, its purchase commitments are what get cut, in cascade. The CDS dispersion in Credit leads equities prices exactly the weakest nodes of this web (CoreWeave, Oracle).