A decade in the pitch, three or four years in the paperwork

Airlines borrow against their planes on the premise that the aircraft will keep flying and earning for decades, so the collateral outlives the loan by a wide margin. Nvidia's $500 billion financing plan proposes the same arrangement for AI chips — developers would pledge high-end GPUs as long-term loan collateral — and the company tells its lenders those chips will keep generating revenue for up to a decade. The paperwork on the other side of the table carries a different number. The lenders are working from a depreciation schedule of three or four years. ^[1] Depreciation is the one place where a belief about an asset's lifespan has to be committed to writing. It is the schedule that assumes the asset sheds value at a set pace and names the point where it is treated as spent — and because a loan has to be repaid before its collateral runs out, the schedule effectively sets the loan's clock. A pitch deck can call a chip durable for free. A term sheet has to date the moment it stops being. Nvidia's own words for what it is selling:

AI compute is a productive, durable and fungible asset that can support long-term financing. — Nvidia

Everyone else gets to assert the decade without signing anything. Nvidia's chief executive has spent the year describing AI as essential national infrastructure, the same category as energy, telecoms and the internet. ^[2] By June the administration had settled the matter from the podium — the Treasury secretary called the country an AI superpower and the outright world leader, and the president put the margin more plainly. ^[3]

The U.S. is the AI leader in the world. — Scott Bessent

The company itself began paying out like a mature utility. It raised its quarterly dividend twenty-five-fold, and in one recent quarter it returned $26 billion to shareholders — more than the $21.3 billion of free cash flow it generated in the same three months. Free cash flow is what is left after a company has paid to run and expand its business; a firm that hands back more than that is drawing down its stored wealth instead of reinvesting it, the habit of a business that believes the heavy spending is behind it. ^[4] Nvidia frames the raise as rewarding shareholders while it remains a high-growth stock ^[5]; the payouts themselves read like the seller of the frontier collecting rent on a frontier it has already built. And it is now, literally, a landlord: through its compute partnerships the company guarantees a minimum of $36 billion in revenue while taking a share of the rental income on the chips it sells. ^[6] The same posture shows up down the chain. OpenAI, the flagship frontier lab, now runs an advertising unit at a $1 billion annualized pace. ^[7] Meta, which spent heavily enough this year that its chief executive flagged renting out any overbuild as the plan, did exactly that in July, selling raw computing power to outside customers. ^[8] The accounts tied to the president moved the same way in miniature: heavy in the suppliers through the spring — Dell up 196 percent, SanDisk up 364 percent — then into dozens of individual dividend-growth stocks in August, and by October a stated rotation away from infrastructure suppliers toward companies that sell the product rather than the machines. ^{[9][10][11][12]} Third parties hold sole authority over those accounts ^[13], so the moves are a mirror, not a message. And yet the physical buildout is not winding down; it is accelerating. Hyperscaler spending on data centers runs at roughly $800 billion this year and is projected to approach $1.3 trillion in 2027 ^[14], funded increasingly with borrowed money as the cash they generate covers a smaller share of the bill. ^[15][16] The story and the balance sheet are moving in opposite directions in the same months: everyone with something to sell is describing a finished national asset, while the asset itself is being built faster than ever, on debt. Which brings the disagreement back to where it has to be priced. The decade sits in the deck; three or four years sit in the schedule. The lenders were never asked to referee the victory narrative — only to set a repayment clock on the collateral. They have already done it: they wrote down the date when the value runs out, and that date is not a decade away.