Details of Nvidia's $500 Billion AI Financing Plan Still Under Negotiation
Nashnova编辑部
Nvidia's $500 billion AI-infrastructure financing plan, announced with six financial partners, remains at the memorandum-of-understanding stage with no deals executed — the heavily marketed "AI as investable asset" narrative is still far from actual lending.
Where does the $500 billion plan actually stand?
Nvidia partnered with BlackRock, Apollo, Goldman Sachs and three other firms to announce the plan, but according to The Information, the parties have only signed a memorandum of understanding — no real transaction has been executed.
CEO Jensen Huang mounted a rare public roadshow — appearing on CNBC alongside executives from all six firms and publishing a blog post framing Nvidia-powered "AI factories" as an "investable asset class."
This means → the plan is currently more narrative marketing than financing reality; substantive details are expected during this week's earnings call.
Why is Broadcom's TPU deal being used as a template?
The market is largely modeling Nvidia's possible structure off Broadcom's $35 billion financing deal announced in June.
That deal's logic: Apollo and BlackRock issued debt → bought Google TPUs (tensor processing units — custom chips designed specifically for AI workloads) designed by Broadcom → leased them to Anthropic, giving the private company access to compute far beyond its own balance sheet.
The key number: Broadcom provided roughly $29 billion in residual-value guarantees, covering about 82% of total debt. In plain terms = Broadcom backstopped over four-fifths of the risk before buyers would commit.
How does Nvidia's risk structure differ from Broadcom's?
Broadcom is now discussing a second deal worth over $60 billion, which may include $30 billion in subordinated bonds — the highest-risk tranche that absorbs losses first but offers the highest returns.
This reflects a narrowing of Broadcom's credit support share — and Nvidia's plan caps residual-value support at just 25%, far below Broadcom's early-deal level of 82%.
This means → Nvidia's structure asks investors to bear significantly more underlying-asset risk, placing a much heavier bet on GPU resale value holding up over time.
Can GPUs stand alone as collateral? What's the core debate?
Nvidia's core thesis: use the GPU's own value as the collateral base for lending, rather than relying on contract guarantees from investment-grade clients like Microsoft or Meta.
Some tranches are expected to mature within five years, repaid quickly from compute-generated cash flow; the GPU's residual value after debt payoff becomes upside.
But the practical challenge is clear: CreditSights analyst Andy Li notes that existing financing deals typically require contract-level credit enhancement from investment-grade customers — smaller AI companies without such backing struggle to access comparable terms.
Can smaller AI companies actually benefit?
Wayne Nelms, founder of compute-data startup Ornn, put it plainly: "Financing thousands of GPU nodes is one thing, but enabling smaller players with thinner balance sheets to access the same financing is a completely different matter."
In plain terms = large customers have contracts and credit ratings that make banks comfortable lending; smaller companies have neither, and the argument that "GPUs are valuable" alone may not convince lenders.
Whether Nvidia's plan can genuinely open a financing channel for smaller AI companies will be the key test of this $500 billion narrative's real-world value.
Content is for reference only, not financial advice.