NVIDIA is building a financial market around AI compute

From chips to financial assets

NVIDIA’s position in the AI economy has long been built around hardware. GPUs and other parts of the company’s technology are used in some of the world’s largest AI systems and data centers.

Now another layer is beginning to emerge.

On August 10, NVIDIA announced memorandums of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR. The aim is to establish financing platforms and dedicated capital pools for AI infrastructure and NVIDIA-based compute, with a goal of mobilizing more than $500 billion in third-party capital over time.

That distinction matters.

This is not a completed $500 billion fund with the money already waiting to be deployed. NVIDIA has not disclosed how much each partner would contribute, the final financing terms or a timetable for when the full amount could be put to work. The final structures still need to be built.

But the direction is becoming clear: AI compute is no longer being treated only as technology that companies purchase. It is increasingly becoming infrastructure that can itself be financed.

Wall Street starts building the machinery

Just four days after NVIDIA’s announcement, the next part of the picture began to emerge.

Reuters reported on August 14 that Goldman Sachs was already in talks with potential investors about the financing structures. US insurance companies, asset managers and banks are expected to form an important part of the investor base. According to the report, Goldman could provide junior capital and private credit through its asset-management operations, while also helping place debt with private credit funds and potentially, at a later stage, in public debt markets.

That makes the $500 billion plan considerably more concrete.

This is no longer simply a group of major financial firms signing memorandums of understanding. We are beginning to see the outlines of how capital could flow from institutional investors, through credit and financing structures, and into data centers and AI compute.

According to Reuters, the ambition is to create an asset-backed market for AI compute in which debt could eventually be traded more like conventional securities. Broader investor participation is also intended to help reduce financing costs.

NVIDIA could become part of the risk structure itself

NVIDIA’s role could extend beyond supplying the hardware.

Jensen Huang has said that NVIDIA has the ability to backstop up to 25 percent of potential deals, equivalent to as much as $125 billion if the full envisioned volume is realized.

That does not mean NVIDIA has already guaranteed $125 billion.

Nor is it a blanket guarantee covering the entire financing platform. The precise terms, which transactions could qualify and when such support might be activated have not yet been publicly defined.

But the statement matters because it brings another question into focus:

What is AI hardware worth if a borrower can no longer pay?

If hardware and compute form part of the collateral or risk support behind the financing, the economic value of that equipment becomes important if a customer defaults. For the model to work, the equipment needs to remain sufficiently valuable, transferable and capable of being redeployed or sold.

Suddenly, the residual value of a GPU matters far beyond the server hall.

This did not begin on August 10

NVIDIA’s new financing initiative is large, but the convergence of AI infrastructure and private capital has been developing for several years.

In 2024, BlackRock, Global Infrastructure Partners, Microsoft and MGX launched what later became the AI Infrastructure Partnership. At the time, the goal was to mobilize $30 billion in equity capital and as much as $100 billion in total investment when debt financing was included. NVIDIA participated from the beginning as a technical adviser.

In March 2025, NVIDIA and xAI formally joined the expanded partnership. Data centers and energy infrastructure were again highlighted as central parts of the AI buildout to be financed.

This is not the same financing structure as the new $500 billion initiative, and the figures should not be combined as if they represented one common pool of capital.

But they point toward the same broader development:

AI is becoming a category of infrastructure that major financial institutions increasingly treat in ways similar to other capital-intensive infrastructure.

When the GPU enters the credit market

That creates opportunities – but also questions that do not normally arise when a company simply buys a few servers with its own money.

How quickly does a GPU lose economic value when a new generation arrives?

What happens to the value of the collateral if demand for a particular type of compute falls?

How heavily must an AI facility actually be utilized for its revenue to cover interest and financing costs?

What happens if energy prices, grid connections or construction costs change the economics?

And who takes the first financial loss if a financed facility fails to generate the cash flows that were expected?

There are not yet public answers to all of those questions.

That does not mean the financing model is wrong. It means understanding how the risk is distributed becomes just as important as understanding the technical capacity.

There is a strong argument for the model

The other side of the argument matters too.

AI infrastructure is extraordinarily expensive to build. If only a handful of the world’s largest technology companies can finance data centers, power infrastructure and advanced compute from their own balance sheets, access to AI capacity could become even more concentrated.

NVIDIA’s model is intended to open larger pools of capital and give AI developers, companies, cloud providers and other customers access to large-scale compute without requiring every customer to finance the entire investment themselves. Lower financing costs and a broader range of investors are part of that ambition.

If AI compute genuinely produces long-term and relatively predictable revenue, financing it in ways similar to other infrastructure may make economic sense.

But there is an important condition:

someone still has to want to buy all of the compute being financed.

The path of the capital is becoming clearer

That may be the most striking part of the development.

We can now begin to follow the path of the capital quite a long way.

Investors provide capital.
Credit is created.
Data centers and AI factories are financed.
NVIDIA technology is installed.
Compute is sold or rented.
Revenue is expected to service the financing.
The capital is expected to generate a return.

If the models develop as planned, even the debt surrounding the infrastructure could itself become an asset bought and sold between investors.

What began with a chip is developing into an entire financial ecosystem.

And the larger that ecosystem becomes, the more important it becomes to distinguish between what has already been financed, what remains a plan and who ultimately carries the risk.

And where is the human being in the equation?

There is also a larger question that the financing model does not fully answer.

AI infrastructure is being built to create productivity, automation and economic value. At the same time, human labor remains the primary source of income for most people and therefore a central part of the same economy in which these AI systems are expected to operate.

NVIDIA and the financial institutions describe in considerable detail how machines, data centers, credit and capital can be connected.

It is much harder to see an equally detailed model for how people will share in the economic gains if AI changes the amount of human labor the economy requires.

That does not mean the outcome must be fewer jobs or greater economic inequality.

But it leads to a question that is becoming increasingly difficult to avoid:

When hundreds of billions of dollars begin to be mobilized for AI’s machines – has the financial system calculated just as carefully for the people who will live in the economy it is building?

NextNet will return to that question in a separate follow-up.

💬 What do you think?

As AI compute begins to be treated as an investable infrastructure asset, which opportunities and risks do you think matter most? And how should the human side of the equation be considered as capital markets build financing models for an increasingly automated economy?