Business

Nvidia signs up six investment firms for platforms aiming to mobilise $500bn in AI compute financing

Nvidia has signed non-binding memorandums of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to build platforms meant to channel more than $500bn of third-party capital into AI infrastructure.

Abstract illustration of an isometric city of data blocks, in graphite grey and orange.
Abstract illustration of an isometric city of data blocks, in graphite grey and orange.

Nvidia announced on 10 August that it has signed memorandums of understanding with six of the largest names in finance: Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR. The aim is to set up independent financing platforms for AI computing infrastructure. Together, the platforms are meant to mobilise more than $500bn of third-party capital “for the buildout of AI infrastructure over time,” according to Nvidia. The agreements are not yet binding, and Nvidia’s announcement says the partnerships “remain subject to execution of the final agreements.”

The deal tries to answer a question that has grown louder as AI spending has climbed: who pays for the hardware, and on what terms?

What was announced

Nvidia says it will work with the six firms to create “dedicated pools of capital at significant scale at attractive rates for NVIDIA customers.” It names those customers as frontier AI labs, enterprises and AI cloud providers. The company calls these the first compute financing platforms of their kind at global scale.

In a blog post published the next day, chief executive Jensen Huang explained what the $500bn figure is and is not. It is the total capital the platforms are designed to attract over time. It is not Nvidia revenue, not a single fund and not a commitment to any one customer. No timetable, fund sizes or split between partners were given.

Each partner described its role in its own terms:

Partner How it framed its role
Apollo Calls modern compute “a scarce, mission-critical asset class” and points to its long-term capital base
BlackRock Extends an existing relationship with Nvidia, including through the AI Infrastructure Partnership
Blackstone Describes itself as a major investor across the Nvidia ecosystem
Brookfield Says compute is becoming a core part of its AI infrastructure strategy
Goldman Sachs Expects to act as both investor and distributor, aiming to “create a market for credit backed by NVIDIA compute”
KKR Builds on Nvidia’s role as a founding investor in KKR’s Helix Digital Infrastructure platform

How much risk Nvidia keeps

The most important detail is in Huang’s blog post. The capital providers will assess each project themselves, looking at the customer, demand, utilisation, cash flow and residual value. Nvidia supplies the technology platform, and the investors make the financing decisions.

Nvidia does keep some exposure. The post says the company “may provide a residual-value support mechanism for up to 25% of an opportunity,” decided project by project. Residual value is what the hardware is still worth at the end of a financing term. In other words, Nvidia may partly protect lenders against the chips being worth less than expected later on. The post describes this support as limited and meant to complement independent underwriting, not replace it. It also calls the level “substantially lower than other compute-financing arrangements.” No further terms were published.

The post asks directly, “Is this circular financing?” The concern is that a supplier helping to fund purchases of its own products can inflate demand. Huang’s answer is that the initiative “is designed to address that concern,” because outside investors take the credit decisions.

The case for compute as an asset class

Nvidia’s argument is that its accelerators can be financed like other long-lived infrastructure, such as aircraft or data centre buildings, because they keep their value and can move between users. The release calls Nvidia compute “fungible and transferable across customers and operators.” The blog post backs this with market figures it cites, though it does not name where they come from. By those figures, rental prices for Nvidia’s H100 chips rose between late 2025 and mid-2026, and the A100, launched in 2020, is still in commercial use.

Huang put the company’s view in one line: “In AI, compute is revenue.”

What it means

Analysis. AI infrastructure has reached a scale where even the largest technology companies are looking past their own balance sheets. Many of the companies that most want Nvidia hardware, including AI labs, newer cloud providers and enterprises, lack the cash flow or credit ratings of the big cloud platforms. Large, standardised financing pools from private-credit and infrastructure investors could widen the range of buyers able to deploy compute at scale. That would also support demand for Nvidia’s products.

The weak point of the model is the one Nvidia highlights itself: residual value. Accelerator generations turn over quickly. Lenders need confidence that chips financed today will still earn rental income several years from now, when newer hardware is available. Nvidia’s offer to support up to a quarter of that risk on some deals shows it believes in the durability of its hardware. It also means the company keeps a share of the downside if that belief proves optimistic.

The partners appear to see the same challenge. KKR’s co-chief executives put it simply in the announcement: “delivery, not ambition, is the hard part.”

What to watch

  • Final agreements. None of the memorandums is binding yet.
  • The first platforms and projects, including which customers draw on them and at what rates.
  • The terms of Nvidia’s residual-value support, and how it shows up in the company’s financial disclosures.
  • Whether Goldman’s aim of a tradable market for compute-backed credit takes shape.

Sources

See All

More on Nvidia

See All

Type to search. ↑ ↓ to navigate, ↵ to open.