Here's how the AI buildout is actually being funded. 👇 A neocloud spins up an SPV. NVIDIA sells GPUs into it. Private credit funds it via a delayed-draw term loan. A compute customer signs a multi-year contract, and those payments fully amortize the debt. Clean. Ring-fenced. Off the parent's balance sheet. But look at the left side of the diagram. That's the part that matters. NVIDIA guarantees a revenue floor over the contract life — and takes revenue-share upside in return. Translation: lenders are no longer underwriting a leveraged neocloud's ability to re-lease depreciating silicon in 2030. They're underwriting NVIDIA's balance sheet. That single feature unlocks billions in private credit. Three things I'd watch: 🔁 Circularity. NVDA sells the chips, may hold equity in the buyer, may hold equity in the end customer, and now floors the revenue. Recognized revenue is increasingly supported by capital and guarantees that NVDA itself provides. That doesn't make the revenue fake, but it does mean revenue quality and the durability of demand are harder to assess from the income statement alone. 📉 Correlation. The floor commitment is an off-balance-sheet-style obligation whose value depends on compute pricing. It's cheap for NVDA in a tight market and expensive in a glut — precisely correlated with when NVDA's core business would also be deteriorating. Worth watching the disclosure in the commitments and contingencies footnote. 🏦 Risk location. The equity tranche is thin and held by neocloud parents; the debt is held by private credit funds and, increasingly, securitized. If utilization or pricing disappoints, first-loss hits neocloud equity, and NVDA's floor is what stands between private credit and impairment. It's the vendor-financing pattern from telecom in 1999-2000, though with a genuinely different feature: the offtake contracts here are largely signed with investment-grade counterparties before the capital is drawn, which was not true of the fiber build. Source: Morgan Stanley Research