The $99 Billion Loop Nvidia Built Around Its Own Customers

There is a number in Nvidia's disclosures that gets less attention than the revenue and more attention than it should: north of $99 billion in equity investments.
That is money Nvidia has put into the companies that buy its products. Some of it is in AI labs with strategic partnerships — Anthropic and OpenAI among them. Some of it is in neoclouds that build capacity and pay Nvidia for hardware. Some of it sits in public positions that show up in the quarterly 13F, and some of it sits in private companies where the valuation is whatever the last round said it was.
In a boom, this looks like genius. You finance the buyer, the buyer buys your product, the revenue compounds, and your equity stake appreciates. In a slowdown, the same structure has a different name, and that name is vendor financing.
The old version of this, and why it ended badly
Telecom equipment makers did this in the late 1990s. Lucent, Cisco, Nortel and Motorola lent money to carriers who then bought their equipment. The revenue was real, the equipment shipped, and the loans were booked as assets. When the carriers could not pay, the equipment makers discovered that their customers' ability to buy had been manufactured by their own balance sheets. Cisco took billions in writedowns. Lucent nearly died.
The structural difference today is that Nvidia is not lending against future revenue from a struggling telecom operator. It is taking equity in companies with access to enormous capital markets and, in several cases, revenue growing faster than anything in the history of software. The structural similarity is that the equity is illiquid, the valuations are marks rather than prices, and the customer's ability to keep buying depends partly on the supplier's willingness to keep investing.
That is a loop. Loops are fine until someone asks what happens at the edge of one.
Why Nvidia does it, in plain terms
Three reasons, all defensible.
First, speed. A neocloud that wants to build a 100,000-GPU campus needs capital before it needs chips. If Nvidia waits for the debt markets to price that risk, the campus gets built two years later, or not at all. Investing accelerates demand that would otherwise be throttled by financing.
Second, lock-in. An equity stake often comes with a supply agreement, a preferred allocation, or a technical partnership around CUDA and networking. It converts a transactional customer into a strategic one.
Third, information. Being an investor gets you board visibility, roadmap alignment, and a seat at the table when the customer decides which accelerator to standardize on. In a market where the difference between winning and losing a design is a two-year cycle, that information is worth more than the return on the equity.
None of these are sinister. But all three make Nvidia's revenue less independent of Nvidia's balance sheet than a simple hardware supplier's would be.

What the third quarter's 13F showed
On November 16, Nvidia filed its quarterly holdings report, and the portfolio moved in ways worth noting.
Total disclosed holdings fell from $63.44 billion to $48.33 billion. The company cut roughly 53.69 million Intel shares, a twenty-five percent reduction, leaving about 161 million shares worth $19.37 billion. It added about 7.71 million SpaceX shares, a 6.3 percent increase, taking that position to roughly 130 million shares worth $19.68 billion — now the largest disclosed holding, marginally ahead of Intel.
SpaceX is a private company. Marking it requires a valuation, and the valuation comes from tender offers and internal rounds rather than a daily market price. That is not improper; it is how private assets are carried. But it does mean that a meaningful chunk of Nvidia's disclosed portfolio is valued by judgment rather than by trading, and judgment is exactly the thing that gets questioned when a company's leadership is in flux.
Note also the direction: out of a public semiconductor position, into a private space and compute company. Whatever the strategic logic, the liquidity profile of the book got worse in a quarter when liquidity was on everyone's mind.
The customer-side version of the same question
Turn the loop around and it looks like this: if a neocloud's largest investor is also its largest supplier, what is its real cost of capital? What happens to its purchasing plans if the supplier's investment thesis changes? And who is underwriting whom?
Analysts asked versions of these questions on the August call. Kress pointed to strategic partnerships and to AI cloud revenue more than tripling year over year, which is a genuinely strong data point. But growth rates are not the same as credit quality, and a tripling from a small base does not tell you what happens in a down cycle.
The other under-discussed item is the $12.9 billion acquisition of Hugging Face, which is reportedly under Department of Justice review. That deal is not vendor financing, but it is the same category of question: Nvidia is buying positions in the layer where developers choose tools, and the antitrust review will decide how far that strategy can go.
Why the leadership change matters here more than anywhere
Here is the thing about deals built on relationships: they do not transfer cleanly.
An equity investment in a lab or a neocloud is usually signed by a small number of people, and the terms that are not in the term sheet — allocation priority, roadmap previews, technical support, the understanding that a call gets returned at 11 p.m. — are personal. A new CEO inherits the contracts and not the relationships. The contracts are enforceable. The rest has to be rebuilt.
That is a real, if unquantifiable, risk to the loop. Nvidia's investment book works because counterparties believe the company will show up for them. An interim CEO with a ninety-day horizon, a board search for a permanent leader, and a shareholder base that just took a 25 percent hit is a harder counterparty to believe in than a founder who has done this for thirty years.
The question to ask on the call
Not "how much have you invested," which is disclosed. Not "what is the return," which is unknowable for private assets.
Ask this: of the $99 billion, how much is committed but not yet funded, and what are the conditions? If a large share is contingent on milestones — capacity delivered, chips purchased, revenue thresholds met — then the loop is more conservative than it looks, and the disclosure would be a comfort. If most of it is unconditional, then Nvidia has effectively written a put option on the AI buildout, and the premium on that option is being paid in equity marks.
Nobody outside the company knows the answer. That is the point. A vendor financing structure is fine as long as everyone can see its shape, and the shape of this one has been disclosed in aggregate rather than in detail.
For thirty-three years, the market accepted that opacity because the person running it had earned the benefit of the doubt. The benefit of the doubt is the thing that left on November 12, and it is not obvious it comes back before the loop gets tested.
Related articles
AMD Got 19 Percent for Being the Other Option
Money did not leave AI. It left the one name with a governance problem and went to the two without one.
The 74 Percent Gross Margin Has a Shelf Life
Three forces arrive in the same four quarters, and each one costs about $3 billion per point.
175,000 Signatures Will Not Put Him Back. Here Is What Would.
The longer the campaign takes to win, the weaker its case becomes. That is the trap built into it.
The Export Rule Moved the Unit of Control From the Chip to the Customer
The rule does not ban chips. It asks who the end user is, which is a much larger regulatory surface.