SANTA CLARA, 16 AUG 2026 — Nvidia disclosed on Friday that it holds 122.8 million SpaceX shares worth about US$21 billion, and roughly US$30 billion of Intel stock. Between them those two positions account for close to 80 per cent of its disclosed equity portfolio.
Both companies have committed to buying Nvidia chips. That is the shape of the story: the largest supplier in the industry now owns large stakes in two of its most important customers.
What the filing shows
The SpaceX holding came out of Nvidia's participation in xAI's US$20 billion funding round in January. The Intel position is reported as an unrealised gain of roughly US$25 billion on an outlay near US$5 billion, with Intel trading around US$140 a share at the end of June.
SpaceX is Nvidia's second-largest disclosed holding, behind Intel.
A private company in a public filing
One detail deserves care, because it is unusual enough that several outlets described the filing as a surprise.
A 13F ordinarily reports holdings in exchange-traded securities. SpaceX is private and its shares do not trade on an exchange, so a line item of that size for a private company is not the routine content of such a filing. We could not establish from the coverage which form carried the disclosure or under what requirement, a point worth stating plainly.
What is not in doubt is the disclosure itself and the figure attached to it.
Why a supplier buys its customers
The strategic logic is not hard to read, and it is not primarily financial.
Nvidia sells accelerators into a market where the binding constraint has moved from demand to supply and where every large buyer is trying to reduce its dependence on one vendor. Google builds its own tensor processors. Amazon has Trainium. Anthropic is reported to be assembling a silicon team. The direction of travel across Nvidia's customer base is towards alternatives.
An equity stake does not prevent that. An equity stake aligns a customer's balance sheet with Nvidia's, gives Nvidia a view of the customer's roadmap, and gives the customer a reason to hesitate before switching suppliers. It is a softer form of lock-in than a contract and a more durable one, because it does not expire.
The Intel position is more complex. Intel is simultaneously a customer, a potential foundry supplier, and a competitor in accelerators. Holding US$30 billion of it while it raises US$20 billion of fresh equity puts Nvidia inside the capital structure of a company whose manufacturing capacity it may eventually need.
The position was received, not bought
The provenance of the SpaceX stake changes how to read it.
Nvidia did not go out and buy shares in a rocket company. It put money into xAI's funding round in January and ended up holding SpaceX equity, which implies the two Musk companies were financially entangled in a way that round made concrete. Nvidia's exposure to SpaceX is therefore partly a consequence of backing a model developer rather than a deliberate decision about launch vehicles or satellite constellations.
That distinction matters for what it predicts. A stake bought deliberately suggests a strategy of acquiring equity in customers. One acquired as a by-product suggests this was a one-off deal with an unusual structure.
Which of those it is will be visible in whether Nvidia does it again, and to whom.
The concentration is the risk
Eighty per cent of a disclosed portfolio in two names would be an unusual concentration for any institution. For a supplier holding stakes in its own customers it is something more specific.
Nvidia's revenue depends on those customers buying chips. Its balance sheet now also depends on their share prices, which depend in part on their ability to buy chips. If AI capital expenditure slows, Nvidia takes the hit twice: once in orders and once in the carrying value of the positions.
That correlation is the opposite of what a corporate treasury normally seeks. It is a strategic bet, not a diversification play.
What this looks like from here
The regional consequence is indirect and worth thinking about anyway.
Southeast Asia's AI infrastructure build-out runs on Nvidia hardware almost without exception — the Indonesian and Malaysian data centre projects, the Indian clusters, the Singapore deployments. Supply allocation in a constrained market is a commercial decision, and a supplier with equity in some customers and not others has an interest that a pure vendor does not.
There is no evidence of preferential allocation and we are not alleging any. The issue is structural. When a supplier is also a shareholder in its largest buyers, it has interests beyond the order book in deciding who gets scarce capacity first — a process regional buyers have no visibility into.
The second consequence is about the shape of the cycle. Nvidia has now tied a substantial part of its balance sheet to the continuation of AI capital spending, on top of a revenue line that already depends on it. This is a bet on continued spending, made by the one company with the best view of the order flow. It signals what Nvidia expects to happen next.
There is one more asymmetry worth naming. Intel is listed, so its position can be marked to a price the whole market can see. SpaceX is not, so its position is marked to an estimate. Two very large numbers sitting side by side in the same disclosure do not carry the same weight of evidence.
What we could not establish
Which SEC form carried the SpaceX line and under what disclosure requirement, given that SpaceX is not exchange-traded and a 13F ordinarily covers listed holdings. Nor how the US$21 billion valuation was arrived at for a private company, which matters because private marks are estimates rather than prices.
Also unestablished: whether Nvidia holds board or information rights in either company, whether the positions are subject to lock-ups, what the remaining 20 per cent of the disclosed portfolio contains, whether any of the SpaceX stake has been sold, and whether the exclusivity arrangements reported for both companies are contractual or a statement of intent.
What to watch
The first thing is whether Nvidia adds to either position. Building further would confirm this as a deliberate strategy of taking equity in customers rather than an artefact of how the xAI round was structured.
Then there is how the Intel relationship develops. Neither company has spelled out the operational logic, but it is not hard to see: Nvidia holds US$30 billion of a firm that is expanding advanced packaging capacity, which is currently the industry's main bottleneck.
Finally, watch whether other suppliers copy it. Taking equity in customers is unusual in semiconductors. If the practice spreads, the line between a supply chain and a web of cross-holdings begins to blur.
