The Chip Company Turned Up on a Rocket Company's Shareholder List

Nvidia filed its Q2 13F on August 14, and there was a name in it nobody had seen before: SpaceX. 122.8 million Class A shares, valued at roughly $21 billion as of June 30.

That makes it Nvidia's second-largest holding. Intel is first, and those two positions alone account for more than 80% of the disclosed equity portfolio. A semiconductor company's investment book is, functionally, two stocks.

Where the stake came from is half the story. Nvidia invested $10 billion in xAI in January 2026 — half of a $20 billion round. Then SpaceX acquired xAI in February at a reported $1.25 trillion valuation, and Nvidia's xAI position converted into 122.8 million SpaceX Class A shares.

SpaceX closed June 30 at $170.86, so 122.8 million shares works out to about $21 billion. Ten billion in, roughly doubled in six months.

Here's the deal, though: don't stop there. SpaceX closed at $140 on August 14. Run the same share count and you get $17.2 billion. The $21 billion in the filing is a photograph of one specific date, and by the day the photograph was published the number had already changed.

How a Chip Company Ended Up Holding This

Nvidia did not used to be this kind of company. A fabless semiconductor designer carrying a multi-tens-of-billions equity portfolio on its books is not normal. But over the past two years Nvidia has repeatedly invested directly into the companies that buy its chips. The capital required to build out AI infrastructure is enormous, and a customer who can't raise it doesn't buy chips.

xAI was the marquee case. Elon Musk founded it in 2023 and built a large GPU cluster in Memphis. From Nvidia's side, xAI was both a major customer and a showcase for its hardware, and January's $10 billion put capital on top of that relationship.

SpaceX's February acquisition of xAI at $1.25 trillion changed the board. Musk's rocket company and AI company merged into one entity, and Nvidia's xAI stake automatically became a SpaceX stake. Nvidia didn't decide to invest in space; the AI company it invested in got absorbed by a rocket company.

And this is where the real consideration surfaces. On SpaceX's earnings call, Musk said Nvidia chips will be the only ones used in SpaceX's AI data centers and frontier AI models. That sentence may be worth more to Nvidia than the $21 billion. What $10 billion bought wasn't just a paper gain — it was a large customer's exclusive supply commitment.

Tech Times read this alongside the Intel position, framing Nvidia's two largest holdings as companies that have both committed to buying its chips.

The Numbers

Item Figure Timing / note
Shares held 122.8M SpaceX Class A End of Q2 2026
Disclosed value ~$21B At June 30 close of $170.86
Portfolio weight ~33.06% Of disclosed equity portfolio
Rank No. 2 (Intel is No. 1) Two positions exceed 80% combined
August 14 revaluation ~$17.2B At $140 close
Original investment $10B into xAI January 2026, part of a $20B round
Conversion event SpaceX acquires xAI February 2026, $1.25T valuation
Side term Nvidia chips exclusive in SpaceX AI data centers Per Musk

The row that matters most is the one where $21 billion sits next to $17.2 billion. A 13F is a quarter-end snapshot published up to 45 days later. The stock fell 18% in between, and the headline number was stale on arrival. That lag is a permanent feature of reading 13Fs; on a volatile position it can be worth $4 billion.

Second, look at that 33%. One position is a third of the portfolio and two positions are over 80%. This isn't a diversified book — it's strategic equity. Nvidia isn't managing assets; it's planting capital in its own ecosystem.

Third: $10 billion in, $21 billion at end of June, $17.2 billion in mid-August. Still a large gain. But it's an unrealized gain, and Nvidia has neither a reason nor an easy structure to sell. Being a major shareholder in a major customer is the point of the investment.

What Each Side Gets

Nvidia gets three things: the paper gain, customer lock-in, and information. A large shareholder sits close enough to see a customer's infrastructure plans early. Knowing an AI data center buildout six months ahead of the market has concrete value for a company planning production.

SpaceX got capital and supply certainty. Having Nvidia as a shareholder improves your position in the GPU allocation queue — and the industry keeps repeating that queue position is harder to get than money. SpaceX pinned its position with equity.

Musk merged two of his companies into one giant entity and put Nvidia on the cap table in the process. Rockets, satellites, and AI in one corporation, tied to its semiconductor supplier by ownership.

Nvidia shareholders got something more ambiguous. They bought a semiconductor company and now hold a book where one large position is 33% of disclosed equities. When SpaceX moves, it lands on Nvidia's income statement as a mark-to-market gain or loss — volatility unrelated to the core business, blended into reported results. Between June 30 and August 14 alone, that was nearly $4 billion of movement.

Rival chip companies lost a market. If SpaceX's AI data centers are Nvidia-only, there's no room for AMD or the custom-silicon camp. And if sealing large customers via equity becomes standard practice, the competitive terms of the whole industry change.

Precedents for Suppliers Buying Their Customers

This pattern recurs in semiconductors and telecom, and results have tracked the cycle.

The successes were the ones where investment and revenue pointed the same direction for a long time. Supplier invests in customer, customer buys equipment with the money, revenue grows, stake appreciates. In an up-cycle the flywheel works for both sides.

The failures showed up on the way down. When demand turns, two things break at once: customer revenue falls, and the value of the stake in that customer falls with it. Risk that should be diversified stacks in one direction instead. Telecom equipment vendors who extended vendor financing to carriers in the early 2000s lived exactly this — revenue evaporating and receivables souring in the same quarter.

Nvidia's version is equity rather than lending, which changes the mechanics somewhat. The correlation is identical, though. If AI infrastructure investment slows, chip revenue and stake value fall together — and with more than 80% of the disclosed portfolio in two names, there's no diversification to soften it.

One more thing: the 18% drawdown between June 30 and August 14 demonstrates this isn't theoretical. Nothing has gone wrong yet, and the position still moved almost $4 billion in under two months.

How Competitors Respond

AMD will likely answer in the same grammar: capital into customers to fix the relationship. You can already see Nvidia and AMD landing in the same AI infrastructure rounds, which signals that equity-as-customer-lock is becoming a standard tactic rather than an exception.

The custom-silicon camp plays a different game entirely. Google with TPUs and Amazon with Trainium run their own chips in their own data centers and sit outside this dynamic. When a company like SpaceX that has to buy chips gets bound to Nvidia, the gap widens between companies that can build silicon and companies that can't.

The big three clouds are spectators and competitors at once. SpaceX building its own AI data centers means SpaceX not renting cloud. Large AI users defecting to owned infrastructure hits cloud revenue directly.

There's a regulatory variable left over. A supplier becoming a major shareholder in a large customer, and that customer committing to buy only that supplier's chips, is the shape of thing competition authorities look at. No inquiry has been reported. But if the structure proliferates, questions are a matter of time.

So What Actually Changes

For Nvidia investors, there's a new way to read earnings. The gap between operating income and net income can widen going forward, because SpaceX's price movement lands as a mark-to-market item mixed in with core performance. Reading the operating line separately is now necessary rather than optional.

For people in AI infrastructure, there's a signal: large GPU allocations increasingly track equity relationships. Showing up with money alone ranks below relationships fixed by ownership or long-term contract. That belongs in procurement strategy.

For founders, there's a structure worth understanding. A hardware supplier like Nvidia or AMD joining your round means more than the money. Just verify whether the consideration includes terms that narrow your freedom of hardware choice later.

For retail investors, there's a lesson about 13Fs. The filing is a quarter-end snapshot published up to 45 days later. Reading the headline number as present value is simply wrong. In this case the lag was worth $4 billion.

For the semiconductor and infrastructure industry generally, the trend is that chip suppliers entering their customers' capital structure is becoming the standard instrument for fixing a relationship. Holding a large customer on a supply contract alone is getting harder.

🥄 Three Things You're Probably Wondering

— Did Nvidia make $21 billion? It's an unrealized gain, not a realized one. And $21 billion is the June 30 figure; recalculated at the August 14 close — the day the filing came out — it's $17.2 billion. Still a large return on $10 billion, but reading the headline number as current value gets you the wrong answer.

— Why not sell? Because there's no reason to. The purpose of the investment was the relationship, not the trade. SpaceX committed to Nvidia chips exclusively in its AI data centers, and the equity is the collateral on that commitment. Selling loosens it.

— Does this affect Nvidia's reported earnings? Yes. Mark-to-market movement on the holding flows into the income statement, blending volatility unrelated to the core business into results — nearly $4 billion of movement between June 30 and August 14 alone. It doesn't change cash flow, though, so it should be read separately from operating performance.

References

Numbers and criteria are as of announcement and may change. Investment calls are yours to make!