Is This an Asset Class or Is This the Same Circle With More People In It

Reinsurance markets learned a hard lesson decades ago. Spreading a risk across many insurers only reduces danger if the underlying event those insurers are all exposed to is itself independent. Ten insurance companies all covering flood damage in the same floodplain have not diversified anything. When the flood comes, all ten face the same claims at the same time, which is precisely the outcome diversification is supposed to prevent. I think Nvidia just built a version of that same structure, and called it something else.

What Was Actually Announced

Nvidia signed memorandums of understanding with six of the largest capital allocators in the world, Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR, to establish independent financing platforms designed to mobilize more than 500 billion dollars in third-party capital for AI infrastructure. Jensen Huang framed this directly as the moment NVIDIA compute becomes an investable asset, productive, revenue generating, and fungible enough to be used across nearly every cloud provider. The structure is designed specifically so this capital does not add to Nvidia's own balance sheet. Outside investors fund the buildout. Nvidia's exposure, at least on paper, goes down.

Here is the mechanism worth understanding before accepting the asset class framing. Borrowers under these platforms are expected to deploy Nvidia specified hardware architectures, which standardizes the collateral and gives lenders a defined recovery path if a borrower defaults. In practice, that means loans collateralized by the GPUs themselves, priced the way a lender might price a loan against a building or a toll road, an asset expected to hold its value over the life of the debt regardless of what happens to any single company involved.

Why That Comparison Does Not Actually Hold

A toll road's value does not depend on whether the company that built it remains dominant in road construction. That independence is what makes it a genuine asset class rather than a bet on one company's continued success. Nvidia's GPUs do not have that same independence, and one analyst flagged exactly why in coverage of the announcement. If Chinese domestic chip production triggers a price war, the value of the GPUs backing these loans could erode quickly, at precisely the moment demand softness would also be causing borrowers to default. Finance has a specific term for that pattern: wrong-way risk. That's when your collateral loses value for the same underlying reason your borrower is failing to pay. It is considered one of the most dangerous risk structures a lender can hold, precisely because it defeats the purpose of collateral in the first place.

The borrower profile makes this sharper still. Reporting on the deal indicates these financing platforms are expected to serve largely non-investment grade borrowers, AI startups and neoclouds locked out of traditional debt markets, with expected yields in the 11 to 17 percent range depending on where an investor sits in the capital structure. That yield range is itself a tell. Real asset classes with genuinely independent cash flows do not typically require double digit returns to attract capital. High yield, high depreciation pricing is the market's own way of saying it does not fully believe the real estate comparison either.

Why This Is the Same Circle, Not a New One

A few weeks ago I wrote about Nvidia's discussions to guarantee financing directly for OpenAI's Ohio campus, a structure where the guarantor and the beneficiary sat on the same side of the transaction. This announcement looks different on the surface, six outside institutions instead of one guarantor, but the underlying exposure has not actually left the building. It has been distributed across six more balance sheets, all of which are now exposed to the same correlated risk, whether AI demand materializes at the scale currently being financed, and whether Nvidia's own competitive position holds long enough for the collateral backing all of it to hold its value.

Recruiting six of the most sophisticated capital allocators on earth into a deal does not, by itself, prove the deal is sound. It proves those six firms were willing to be recruited, at a price, into the same underlying bet everyone else in this space has already made. Sophistication of the counterparty is not the same thing as independence of the risk.

What This Means for Anyone Evaluating This as a Genuine Asset Class

Run the actual test before accepting the framing. Ask whether the cash flows backing a specific financing structure would hold up if the company promoting it disappeared tomorrow, the way a toll road's revenue would. If the honest answer depends on Nvidia's continued market dominance, continued chip export access, and continued demand at the scale currently being financed, you are not looking at a new asset class. You are looking at the same concentrated bet, now underwritten by more institutions who each have their own reasons for wanting to believe otherwise.

If you want to run a specific financing structure through that same test, tracing whether its cash flows are genuinely independent of the parties promoting it, I would rather do that with you before "investable asset class" gets treated as a settled description in your own underwriting.

If that's where you are, whether you're walking through this yourself or bringing it to a board, grab 15 minutes on my calendar and let's set up time to go through it.

PLUS:

Grab the guide. Before advancing a data center site, there are 12 questions I make sure I can answer. I wrote them up here: [The 12 Questions Every Real Estate Professional Should Ask Before Advancing a Data Center Site].