Nvidia’s quarterly filing dropped a surprise this month: the company holds about 122.8 million Class A shares of SpaceX, worth roughly $21 billion at the end of Q2 2026. It’s the second largest position in Nvidia’s equity portfolio, at around a third of it, and the paper trail shows the stake was never bought directly.
Where the stake came from
The origin is a chain of deals. In January 2026, Nvidia put $10 billion into xAI. A month later, SpaceX acquired xAI, and Nvidia’s xAI shares converted into SpaceX equity. SpaceX’s private-market valuation did the rest, turning $10 billion into $21 billion on paper by June 30.
The position came to light through Nvidia’s Form 13F filing on August 14. SpaceX shares have moved since, and the valuation has swung with them: when the price dropped from about $170 to $140, the holding fell to roughly $17 billion. That kind of volatility now flows through Nvidia’s other income line, which analysts have flagged as a new source of earnings noise.
The deal inside the deal
Alongside the stake, Musk has said SpaceX will use Nvidia’s Vera Rubin GPUs exclusively for its AI data centers, and SpaceX plans to grow its computing capacity from about 2 gigawatts to close to 10 gigawatts by the end of 2027. That’s the pattern worth paying attention to. Nvidia has committed more than $100 billion to AI companies over the past two years, including CoreWeave and Cursor, and the structure is usually the same: Nvidia invests, the customer buys Nvidia chips.
Critics have called this vendor financing, and the label isn’t unfair. It echoes the telecom equipment deals of the late 1990s, when suppliers funded their own buyers. The defense is that Nvidia’s money is chasing real demand and its investments are performing. The 2x paper return on xAI within months supports that defense. The risk is concentration and circularity: Nvidia’s revenue depends on customers who depend partly on Nvidia’s capital, and a slowdown anywhere in that loop hits twice.
Accounting treatment adds another wrinkle. Positions like this sit in Nvidia’s other income line, which means a private-company markdown shows up in earnings even when the core business is fine. Analysts who normally model Nvidia on data center revenue now need a view on private-market valuations for SpaceX and friends. Investors got a preview of this dynamic with earlier stakes in public companies, but a holding this large in a private company with no market price makes mark-to-market guesswork part of the quarterly report.
It cuts both ways, of course. The same mechanism that produces noise on the way down produced a windfall on the way up, and no shareholder complained about the $11 billion of appreciation. The question is whether Nvidia’s core earnings get drowned out by its investment portfolio if more of these mega-stakes land. For a company whose valuation rests on being the cleanest read on AI demand, an income statement that also moves with rocket-adjacent private markets is an odd thing to explain every quarter.
Why SpaceX
SpaceX is an unusual customer for a GPU vendor, but the logic connects through compute. The company is building out AI data centers on the ground and has talked publicly about orbital datacenters and Starlink’s expansion. Ten gigawatts is utility-scale, the kind of number that buys a lot of accelerators, and Musk committing to a single GPU vendor makes SpaceX one of Nvidia’s most important forward customers. The equity stake aligns both sides: Nvidia gets upside in the customer it’s supplying, and SpaceX gets a supplier with a reason to keep its roadmap pointed at the company.
There’s also a competitive angle. SpaceX is one of the few companies with its own power generation ambitions, launch cadence, and network infrastructure, meaning it can credibly build data centers where grid connections are the bottleneck for everyone else. xAI brought its Colossus-class training ambitions into the merger, and the combined entity has reasons to want guaranteed GPU allocation rather than queueing behind hyperscalers on Nvidia’s allocation list. Owning a chunk of your supplier’s best customer, and having that customer own a reason to favor you, is one way to skip the queue.
What developers should take from it
For builders, two things follow. First, the shape of AI infrastructure keeps shifting toward vertically integrated players, and a chip vendor with equity in a launch provider is a new configuration. If SpaceX’s compute plans materialize, some of that capacity will presumably be resold in some form, and Starlink’s network already reaches places terrestrial fiber doesn’t. Second, Nvidia’s filings are now worth reading the way you’d read a hyperscaler’s capex guidance. The company’s balance sheet has become a map of who’s buying AI compute, and the 13F is where that map shows up first.
The open question is regulatory. Equity stakes in customers that come with purchase commitments draw antitrust attention, and the bigger these positions get, the harder it is to argue the investments are passive. Nothing has been announced on that front, and Nvidia didn’t respond to requests for comment on the filing. Watch the next quarterly disclosure: whether the SpaceX position grows, shrinks, or gets restructured will say a lot about how management views the circularity question.