Nvidia lines up six Wall Street firms to source $500 billion in AI financing — its second unrelated $500 billion deal since July

[Nvidia](#/company/nvidia) said on August 10 that it has entered non-binding memoranda of understanding with six of the world's largest private-equity and asset-management firms — Apollo Global Management, Blackstone, BlackRock, Brookfield Asset Management, Goldman Sachs and KKR — to source more than $500 billion in third-party financing for AI infrastructure. CEO Jensen Huang told CNBC he approached only those six firms for the commitment, and none turned him down.

What the deal actually does

The structure is financing platforms, not a single fund: special-purpose vehicles will be able to issue tens of billions of dollars in debt at once — through private placements and bonds — then lease Nvidia compute to buyers, with Goldman Sachs positioned as lead bookrunner on the public debt side. Nvidia itself may put in up to 25% of the financing on any given deal. Huang's pitch, in his own words, is that these platforms "will help customers access scarce compute at scale," treating GPU clusters less like fast-depreciating tech hardware and more like a long-term, revenue-generating asset — the same category as real estate or a power grid, in Huang's framing. BlackRock CEO Larry Fink backed that read, calling the resulting debt "high credit quality" with "attractive yields" for investors otherwise heavy in equities.

The timing is the context that makes this land as it does. This is Nvidia's third distinct headline financing or infrastructure figure disclosed in about three weeks, after [the SK Group buildout](#/article/nvidia-sk-group-korea-500-billion-ai-partnership) and [the reported Ohio chip-financing and lease-guarantee figures](#/article/nvidia-openai-250-billion-ohio-data-center-financing) — each structured differently, each announced within days of the last. That pace is itself part of the story: whether it reflects demand outrunning any single financing channel, or a company reaching for every available lever at once, is exactly the question the skeptical read below raises.

Reading the two side by side matters because both are being cited in the same week as evidence of AI-infrastructure scale, and they measure almost nothing in common — one is a country-scale build-out with equity stakes attached, the other is a debt-financing capacity target with no infrastructure of its own yet. Against the wider pattern, this financing alliance is Nvidia's latest entry: prior figures on the books include [a reported $350 billion in chip financing and a $250 billion lease guarantee tied to OpenAI's Ohio campus](#/article/nvidia-openai-250-billion-ohio-data-center-financing), each a separate instrument again.

The case against it

The skeptical read starts with what the MOUs are not: binding contracts. Nothing in the $500 billion figure is guaranteed capital today, and the platforms themselves are still to be built and marketed over the coming months. Hedgeye's Felix Wang put the mechanism plainly — by financing customers' access to its own chips, Nvidia has "effectively made Nvidia's product cheaper without really cutting GPU prices." That's a specific version of the circular-financing critique that's followed Nvidia's other megadeals: routing capital through vehicles that ultimately buy more Nvidia hardware doesn't manufacture new underlying demand, even when every dollar involved is real.

Not every read is skeptical. Zuma Wealth's chief investment officer, Terri Spath, told Fortune she's "very bullish on the earnings power of AI" even while cautioning that current valuation multiples leave little room for error. That split — confidence in the underlying compute demand alongside caution on how it's being financed and priced — runs through most of the reaction to this deal, rather than a clean bull-versus-bear divide. The financing structure and the demand it's meant to serve are two different questions, and this announcement mostly answers the first.

None of that makes the underlying compute demand fictional. Nvidia's arrangement with SK Group three weeks earlier is already funding a physical 2-gigawatt data center and real HBM4 production lines, regardless of how this second, unrelated $500 billion is eventually drawn down. What the skeptical case actually targets is narrower: whether financing engineered to make Nvidia's own hardware easier to buy is evidence of durable end demand, or evidence that the industry is running out of investors willing to fund AI infrastructure the ordinary way.