NVIDIA Partners with Six Major Financial Institutions to Launch $50 Billion Compute Asset-Backed Financing
Nashnova编辑部
Nvidia is working with Apollo, BlackRock, and four other financial heavyweights on a $50 billion financing plan to package GPUs as investable infrastructure assets — the AI race now extends from chip performance to who can secure cheaper capital.
Turning chips into "financial assets" — what does that actually mean?
The core idea: Nvidia wants GPUs treated like power plants or highways — infrastructure assets that financial institutions will hold and lend against.
This means → the AI computing race has a new dimension: it is no longer just about who has the fastest or cheapest chip, but about who gets the best financing terms.
CEO Jensen Huang calls it "the first time tech chips have truly become an investable asset class." BlackRock CEO Larry Fink compared it to the birth of mortgage-backed securities in the 1970s — bundling home loans into tradeable products — calling it "the next future of financial engineering."
Why should a GPU qualify as an "asset"? Does Nvidia's case hold up?
Nvidia's argument: GPUs run on a massive software ecosystem (CUDA) and are not locked to one customer. If the original tenant exits, the hardware can be redeployed, re-leased, or resold to other cloud providers — giving GPUs a credible secondary-market value.
But Huang himself said at Nvidia's AI summit that once Blackwell ships at scale, "you couldn't give Hopper chips away."
In plain terms = claiming chips are "long-lived and fungible" while also saying last-gen hardware is worthless creates an obvious tension. Investors need to decide which statement is closer to reality.
Broadcom already moved first — how is its structure different?
Broadcom's XPV structure is already live at roughly $35 billion. A special-purpose vehicle — essentially a shell company set up for one deal — issues debt, buys custom XPU chips from Broadcom, then leases them long-term to Anthropic and repays bondholders with the rental income.
The key difference is the depth of the residual-value guarantee (RVG): if the tenant defaults and asset-disposal proceeds fall short, Broadcom may have to cover the gap. The first deal reportedly puts about $30 billion in senior debt within the scope of that guarantee.
Nvidia's exposure is theoretically lighter: residual-value support of up to roughly 25% kicks in only if disposal prices miss targets and specific contract triggers are met. But neither the triggers nor the legal enforceability have been fully disclosed.
Where is the biggest risk?
Former hedge-fund manager Mark Rubinstein warns that MBS collapsed when mortgage origination became excessive. Two parallel pressures exist today: data-center supply keeps expanding, and Chinese open-source models are already competitive at lower compute requirements — both challenge the core assumption that compute demand only goes up.
Whether frontier AI labs — Anthropic, OpenAI, and others — can actually turn a profit remains an open question. If the tenants themselves are not profitable, the reliability of rental cash flows is in doubt.
This reflects a deeper issue: what determines the real risk is not whether the financing sits on- or off-balance-sheet, but how much of the chipmaker's own credit is embedded in the project debt.
Is the market already pricing this risk transfer?
The signal is visible: CDS spreads — the price of insuring against default — for the big cloud hyperscalers are tightening, while CDS spreads for Nvidia and Broadcom are widening.
This means → by providing credit enhancement and residual-value backstops, upstream chipmakers are shifting part of the project risk onto their own credit — and the market is repricing accordingly.
The entire $50 billion plan remains at the memorandum-of-understanding (MOU) stage with no binding contracts. Nvidia signed a $1 billion MOU with OpenAI last year that never materialized — a precedent worth keeping in mind.
Content is for reference only, not financial advice.