Elon Musk said this week that Nvidia and SpaceX have designed hardware to put an AI data center into orbit, posting on X that the companies "designed a space-optimized Vera Rubin NVL72 system for launch to orbit in Q4 next year, with significant scale in 2028." The post, made in reply to an Nvidia announcement about its new Vera CPU, is the most specific public timeline either company has given for what Nvidia and Musk have been calling the SpaceX Starmind AI satellite project. [1]
The relationship between Nvidia chief executive Jensen Huang and Musk has been tightening for months, and Wednesday's earnings call gave the clearest picture yet of how far it now extends. Nvidia disclosed a $21 billion stake in SpaceX earlier this month, the product of an earlier $10 billion investment in xAI ahead of that company's merger with Musk's rocket firm; those shares are now worth closer to $17 billion after SpaceX's stock cooled from its record June IPO. [2] On SpaceX's own first earnings call as a public company, Musk went further than the equity relationship, endorsing Nvidia's Vera Rubin platform outright and saying SpaceX would build its AI data centers exclusively using Nvidia chips. [1] Musk added on that call that he expects a "significant" allocation of Vera Rubin chips next year, though neither company has said how many chips the orbital plan will ultimately require. [1]
The orbital ambition is an extension of, not a departure from, that exclusivity commitment. Nvidia's own post that Musk replied to described the Vera CPU as built "for agents," designed to handle "the orchestration, code execution, and data processing that powers" SpaceX's next generation of AI systems — language aimed at terrestrial data centers as much as orbital ones. Musk's reply narrowed that general pitch into a specific hardware configuration and a specific quarter: a space-hardened NVL72 rack system, Nvidia's dense GPU server architecture, adapted to survive launch and operate in orbit starting in the last three months of 2027.
What makes this a genuine case of X outrunning mainstream coverage is sequencing. Musk's post went up Monday, August 24, and CNBC's earnings dispatch two days later folded the orbital detail into a section on the broader Huang-Musk relationship — useful context, but secondary to the quarter's headline numbers on revenue and guidance. [1] No outlet has yet published a standalone story on the orbital data center plan itself, what it would take to cool GPU racks in vacuum, or how Nvidia's supply chain — already strained enough that the company's own filings this week flagged rising indebtedness as a new risk factor — would absorb a space-rated production run alongside its terrestrial commitments. [1] For now, the most detailed account of a genuinely novel infrastructure plan exists in a 280-character post, not a press release.
The Nvidia post Musk replied to was itself part of a broader product push. Nvidia is currently ramping up sales of Vera, its first ground-up CPU design, positioned to compete directly with the traditional CPU leaders Intel and AMD. [1] Huang told CNBC's Jim Cramer this week that Amazon Web Services, which is separately buying 2 million Nvidia GPUs, will also purchase "millions of CPUs" built on the Vera architecture, and that some of those chips will be "integrated with Rubin," Nvidia's forthcoming flagship AI chip. [1] The orbital plan Musk described uses the combined Vera Rubin platform, meaning the same hardware generation Nvidia is selling to its largest terrestrial customers is what SpaceX intends to fly. That shared platform is the practical mechanism behind Musk's exclusivity claim: rather than developing bespoke space-rated silicon, SpaceX is betting that Nvidia's standard high-volume product line can be adapted for orbit quickly enough to hit a Q4 2027 launch window.
Nvidia's investment gains from the Musk relationship are already showing up on its income statement. The company booked a $7.8 billion gain on equity investments this quarter, following a $15.9 billion gain the quarter before, with SpaceX and Intel named as the largest holdings driving those numbers. [1] Whatever the orbital data center project ultimately costs to build, Nvidia is, for now, being paid to wait for it.
The financial relationship runs in both directions in ways that make the orbital announcement harder to read as pure hardware ambition. Nvidia's $21 billion SpaceX stake originated as a $10 billion investment in xAI, ahead of that company's merger into Musk's rocket firm — meaning Nvidia became a SpaceX shareholder through a corporate restructuring rather than a direct bet on rocketry or satellites. That history matters for how skeptically to read Musk's exclusivity pledge: a chip supplier that is also a significant equity holder in its customer has every incentive to see that customer commit publicly and specifically to its hardware, and Musk's post does exactly that, on a platform he also owns. None of that makes the underlying technical claim false, but it does mean the orbital data center announcement arrived from a source with a direct financial stake in Nvidia's stock price reacting well to it — a detail that sits alongside, rather than replaces, the genuine novelty of the engineering claim itself.
-- THEO KAPLAN, San Francisco