Nvidia has disclosed that it owned nearly 123 million shares in SpaceX worth almost $21 billion at the end of June, revealing a far deeper financial relationship with one of its largest prospective infrastructure customers than previously known.

The disclosure, reported from a Friday SEC filing by Ars Technica AI, comes shortly after SpaceX chief executive Elon Musk said the company would build its data centers exclusively on Nvidia technology. It also follows SpaceX’s public listing and the combination of Musk’s xAI artificial-intelligence business with SpaceX.

The stake links Nvidia to a customer that is planning a major expansion of computing capacity while raising broader questions about the chipmaker’s increasingly complex role in the AI market. Nvidia is not only selling the processors and networking systems needed to build AI data centers; it is also investing in the companies buying or deploying them and helping arrange financing for infrastructure projects.

What Nvidia’s filing reveals

According to Ars Technica AI’s account of the filing, Nvidia held almost 123 million SpaceX shares valued at approximately $21 billion on June 30. SpaceX shares have since fallen sharply from their value around the company’s initial public offering, meaning the position would be worth about $17 billion based on the report’s estimate.

The filing follows Nvidia’s investment in xAI, completed in January. Musk later combined xAI with SpaceX, creating a direct connection between Nvidia’s earlier AI-lab investment and its newly disclosed ownership of the rocket and satellite company.

The precise terms and timing of Nvidia’s SpaceX investment were not detailed in the supplied reporting. Nvidia did not immediately respond to Ars Technica AI’s request for comment, so the filing is the clearest evidence available in this report of the size of the holding.

Business Chief and Mobile World Live separately carried headlines reporting the same $21 billion SpaceX holding, but their full article text was unavailable in the source material. Those items therefore provide corroboration of the central news event, not additional detail about the transaction or Nvidia’s intentions.

A customer relationship with financial ties

Musk said during SpaceX’s first public earnings call that the company had chosen an exclusive relationship with Nvidia for its data centers. He cited Nvidia’s Vera Rubin architecture as the best option and described the relationship as a close partnership on multiple levels.

That statement matters because SpaceX is preparing to scale its computing footprint rapidly. Musk told investors that the company expected capacity to rise from 2 gigawatts at the end of this year to closer to 10 gigawatts than 5 gigawatts by the end of 2027. The comment was an executive forecast, not an independently verified deployment figure.

For Nvidia, a large ownership position in SpaceX creates an unusually direct connection between capital allocation and hardware demand. Nvidia can benefit if SpaceX expands its use of AI infrastructure, but its financial exposure also means that the value of the relationship is not limited to product sales.

The arrangement adds to a pattern already visible elsewhere in the sector. Nvidia has committed more than $100 billion to AI companies over the past two years, according to the reporting, including investments in cloud-computing provider CoreWeave and AI labs such as Thinking Machines and Safe Superintelligence. Nvidia has also invested in Cursor, the code-editing startup that SpaceX acquired for $60 billion this week, according to the same report.

Evidence, claims, and unanswered questions

The $21 billion figure is based on Nvidia’s SEC disclosure, making it the strongest confirmed fact in the available evidence. The estimated decline to roughly $17 billion reflects the reported fall in SpaceX’s share price rather than a new valuation disclosed by Nvidia.

The exclusive Nvidia arrangement comes from Musk’s public comments and should be treated as SpaceX’s stated procurement position. The statement does not establish how many chips, servers, or complete systems SpaceX has ordered, nor does it confirm that all planned computing capacity will be operational on the stated timetable.

Nvidia also disclosed plans to assemble more than $500 billion from investors including Apollo Global, Blackstone, BlackRock, Brookfield Asset Management, Goldman Sachs, and KKR to help finance customers. The company plans to partially guarantee loans backed by the value of Nvidia chips, according to the report. That proposal was described as a financing plan, not as completed funding or proof of customer deployment.

These distinctions are important for builders and investors. Ownership, announced procurement, financing commitments, and installed compute capacity are separate milestones. The SpaceX disclosure confirms the first and Musk’s comments indicate the second, while the scale and timing of actual deployment remain less certain.

Why the holding matters to AI builders and enterprises

For AI infrastructure teams, the development highlights how hardware availability, capital, and platform choice are becoming interconnected. A supplier that also invests in customers may be able to help them secure equipment and funding, especially when large data-center projects require substantial upfront spending.

That model could accelerate the construction of AI infrastructure. Nvidia’s financing strategy may reduce the immediate burden on customers, while its equity investments can strengthen commercial relationships around its hardware and software ecosystem. For startups and enterprise buyers, however, the model also raises questions about concentration and independence.

A customer that relies exclusively on one supplier may gain access to a tightly integrated platform, but it may have less negotiating leverage and fewer options if product road maps, pricing, power requirements, or delivery schedules change. Nvidia’s financial interests could also make it harder for outside observers to distinguish ordinary customer demand from demand supported by strategic investment or credit arrangements.

The SpaceX case is especially significant because the company’s planned compute expansion is large and connected to several AI businesses controlled by Musk. Nvidia’s stake in SpaceX, earlier investment in xAI, and investment in Cursor place the chipmaker inside a network spanning models, software, data centers, and aerospace operations. The available evidence does not show that Nvidia controls those businesses, but it does show how closely its commercial and financial interests can overlap with them.

What to watch next

The next important signals will be Nvidia’s fuller financial disclosures, including any information about the investment’s purchase price, accounting treatment, and changes in value. Investors will also be watching whether Nvidia comments publicly on the SpaceX relationship or explains how the stake fits within its broader investment strategy.

On the operating side, SpaceX’s actual data-center buildout will matter more than the announced target. Evidence of new facilities, power procurement, hardware deliveries, and deployed capacity would help determine whether the planned increase from 2 gigawatts to as much as 10 gigawatts is progressing.

The market should also track whether Nvidia’s proposed customer-financing structure closes transactions and whether other major AI infrastructure buyers accept similar arrangements. Those developments would indicate whether Nvidia is creating a repeatable financing channel or pursuing a small number of highly strategic relationships.

Creati.ai perspective

Nvidia’s SpaceX disclosure is important not simply because the stake is large, but because it makes the AI supply chain’s financial structure more visible. The leading chip supplier is increasingly acting as investor, lender, and ecosystem partner at the same time it remains a critical vendor.

That can speed deployment for companies with ambitious compute plans, but it increases the need for clear disclosures and careful procurement analysis. AI builders and enterprise buyers should evaluate not only benchmark performance and hardware access, but also ownership ties, financing conditions, supplier dependence, and the difference between announced capacity and working systems.