Nvidia has recruited six of the largest names in finance to help turn its chips into something a bank can lend against.

The company announced partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to build what it calls “compute financing platforms”, structures designed to mobilise over $500bn of third-party capital for AI infrastructure.

The idea is to create dedicated pools of capital at scale, and at attractive rates, for Nvidia’s customers, meaning the frontier AI labs, enterprises and cloud operators buying its hardware.

In effect, Nvidia is helping to build a financing pipeline around its own product, which is why the move reads as a more formal turn in the circular financing binding the AI economy together.

At the centre of the pitch is a reframing of what a GPU is. Nvidia is presenting its compute as “an investable asset”, one it says carries the lowest token cost, the highest revenue, the longest operational life and a rich CUDA-based software ecosystem.

The argument is that a graphics processor is no longer just kit that depreciates in a rack, but collateral with a predictable return, so lenders can treat it much as they would a toll road or a power plant.

Chief executive Jensen Huang made the case in similar terms. “Nvidia compute is uniquely suited for this role,” he said. *“It is broadly adopted, flexible across models and workloads, fungible and transferable.” *

Goldman Sachs, for its part, described its role as “creating a market for credit backed by Nvidia compute”, language that makes the ambition plain: a tradable asset class with the chips as the underlying security.

The partners each bring a different slice of capital. Apollo offers a flexible long-term base, BlackRock connects long-term money to essential infrastructure, KKR combines long-duration capital with infrastructure expertise, and Brookfield, whose reach already extends to a $100bn data campus, would scale the “AI factories” that house the hardware.

No individual project names or amounts were disclosed, and the $500bn figure is an aggregate potential over time rather than a committed sum.

That caveat matters, because so far only memorandums of understanding have been signed.

Final agreements are still pending, which means the headline number describes an intention rather than money that has actually changed hands, and MOUs of this kind can quietly shrink or stall in the long gap between announcement and closing.

The structure is striking for another reason. It formalises a pattern that has already made investors uneasy, since Nvidia sits on multiple sides of these arrangements: it sells the chips, vouches for their resale value, and now helps assemble the capital to buy them.

It is worth remembering that Nvidia’s own $750bn of AI deals rattled its credit market earlier, a sign that even the company’s backers are alert to how tightly these commitments are wound.

The deeper worry is leverage. Most of the half a trillion dollars in view would be debt, and layering it onto an infrastructure boom is exactly the dynamic regulators have flagged.

Indeed, the BIS has warned an AI bust could hit credit markets as hard as 2008, precisely because so much of the buildout now rests on borrowed money and interlocking promises.

And “compute as collateral” only holds while demand does. If AI revenue softens, the asset underpinning all this credit could reprice quickly, leaving leveraged buyers exposed and lenders holding chips worth less than the loans against them.

Huang calls the hardware fungible and transferable, though a glut would test how fungible it really is.

For now, Nvidia has done something subtle but consequential. It has enlisted Wall Street to underwrite the demand for its own products, and if the agreements firm up, the AI buildout gains a vast new source of fuel, even as the debt beneath it grows harder to see through.

Get the TNW newsletter

Get the most important tech news in your inbox each week.