Nvidia's $500 Billion Rewrite: Compute Stops Being Capex and Becomes Someone Else's Debt

Nvidia said Monday it has formed partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to create financing platforms intended to mobilize more than $500 billion for AI infrastructure, with the money coming largely from third-party investors ^[1][2]. That is an accounting and ownership change more than a technology one: GPU buying moves off corporate balance sheets and into vehicles whose debt is held by institutions with decades-long liabilities ^[2][7][8].

The mechanism is not complicated. An independent financing vehicle raises capital to buy Nvidia GPUs and data-center infrastructure, an AI company leases that compute or commits to using it, and the resulting payment stream becomes collateral the vehicle can borrow against ^[6]. Apollo, KKR and their peers structure and manage that debt and place it with the pools of money they oversee, which are mostly insurance and retirement capital ^[7]. Insurers and pension funds want long-duration assets because annuities and retirement benefits are owed decades out, and data centers are long-lived projects needing financing over many years ^[8].

Jensen Huang's framing, published Tuesday, is that the industry has moved "from an era in which companies bought chips and built data centers project by project to one in which AI factories can be financed as productive infrastructure," on the premise that "in AI, compute is revenue" ^[4][5]. Analysts had been waiting for exactly this: a structure that treats AI compute as an infrastructure asset like a toll road or a power plant, producing cash flows that can support debt, rather than a rapidly depreciating pile of graphics processors requiring ever more capital ^[3].

The distinction is who eats the depreciation if the second description turns out to be the right one. Huang said Nvidia may provide residual-value support of up to 25% on some projects, a partial promise against the chips being worth far less than lenders assumed ^[9]. On projects where that support applies at the cap, up to three quarters of the residual exposure still sits with investors ^[10]. Ben Thompson of Stratechery reads the support as "in a certain sense, a price cut," with Nvidia spending its own profits to lower customers' cost of capital and make Nvidia-based data centers easier to finance ^[11].

Sell-side reaction has focused on the circularity question rather than the duration one. Morgan Stanley's Joseph Moore said the arrangement alleviates circular-financing concerns because third-party investors supply most of the capital and Nvidia participates only to a limited extent ^[12]. Bank of America's Vivek Arya broadly agreed and argued Nvidia's chips are unusually financeable because GPUs can be moved between operators, citing CUDA as a supporting factor ^[13].

Context on scale: Goldman Sachs estimates AI-related financing now accounts for nearly a quarter of all gross U.S. investment-grade issuance, with AI investment approaching $600 billion this year ^[14]. The announced $500 billion target is roughly 83% of a single year of that investment ^[15]. Terms, including the size of each individual deal, have not been disclosed ^[16].