Musk Says Nvidia Data Centers Hit Orbit in 2027. SpaceX’s Own IPO Filing Said 2028 at the Earliest
Elon Musk just promised Nvidia data centers in orbit by 2027, but SpaceX's own SEC filing told a very different story to regulators just three months ago, and the gap between those two timelines matters a lot if you own…
Elon Musk posted on X this week that “SpaceX, in partnership with Nvidia, has designed a space-optimized Vera Rubin NVL72 system for launch to orbit in Q4 next year, with significant scale in 2028.” He also called the design “significantly simpler, lower cost, denser and lighter than a traditional rack.”
SpaceX, in partnership with Nvidia, has designed a space-optimized Vera Rubin NVL72 system for launch to orbit in Q4 next year, with significant scale in 2028 https://t.co/qdDq8YBkzl
— Elon Musk (@elonmusk) August 24, 2026
That schedule does not match the one his own company gave regulators. SpaceX (NASDAQ:SPCX | SPCX Price Prediction) filed its Form S-1 on May 20, 2026, and the filing said the orbital data-center program relied on “unproven technologies, or technologies that do not exist,” with deployment projected “as early as 2028.”
The counterparty on the compute side is NVIDIA (NASDAQ:NVDA), whose Vera Rubin platform is the piece Musk wants in orbit. Both stocks closed higher on August 25, 2026, with SpaceX at $137.95, up 2.19%, and Nvidia at $213.05, also up 2.19%.
Two Timelines From the Same Company
Musk named the first satellite Starmind AI1, targeted for the fourth quarter of 2027. It is designed to carry an optimized NVIDIA Vera Rubin NVL72 system rated for 175 kW of average compute and up to 250 kW at peak.
The S-1 language, signed by executives three months earlier, described the same effort in the register regulators expect. The filing classified the company under SIC code 7370, “Services, Computer Programming, Data Processing,” a computing classification rather than an aerospace one, revealing how management wants investors to think about the business.
Companies routinely describe risk conservatively for the SEC and ambitiously on social media. An investor pricing SPCX today is pricing a schedule, and the two schedules differ by at least a year.
SpaceX also announced a $100 billion Louisiana spaceport, with construction expected to begin in 2027 and a first launch targeted for as early as 2029, which comes after both the Starmind AI1 date Musk gave and the 2028 scale target.
Engineering Case and What Heat Does to It
The power argument for orbit is real. A satellite in the right sun-synchronous orbit gets near-continuous solar exposure, faces no grid interconnection queue, and avoids local permitting fights over land and cooling water. Those are the same constraints pushing terrestrial AI buildout into a smaller group of power, cooling, and grid suppliers, which we profiled in a free report on seven AI infrastructure names that aren’t chipmakers.
Thermal load is where that case gets harder. An orbital data center has no ambient air or water to carry away heat, so every watt of waste heat must leave as radiation, and radiating hundreds of kilowatts from a spacecraft has never been done. Reporting on the AI1 design describes roughly 75-meter solar wings paired with 30-meter radiator panels.
Musk told Dwarkesh Patel in February that “you want to design it to be more radiation tolerant and run at a higher temperature,” adding that increasing the operating temperature by roughly 20°K can cut radiator mass in half.
That claim only pays off if Nvidia silicon can be qualified to run reliably at those temperatures for years in vacuum, under proton and heavy-ion bombardment, without the maintenance access a terrestrial hyperscaler takes for granted. Nvidia’s Vera Rubin ramp is scheduled to begin with “production shipments of Vera Rubin in the second half of this year, starting in Q3,” a terrestrial roadmap that has not been qualified for space.
Regulators, Rivals, and What Would Confirm 2027
In February, the FCC accepted SpaceX’s application for filing a constellation of up to one million orbital data-center satellites. Acceptance for review does not amount to approval, and the docket drew roughly 1,500 comments in objection.
The American Astronomical Society petitioned to deny, citing visible-light, infrared, and radio interference, atmospheric reentry effects, and collision risk. Those concerns have already reshaped Starlink’s operational rules and would likely shape any orbital compute constellation.
Amazon’s Leo unit also asked the FCC to refuse the application as speculative, arguing that sustaining a million-satellite constellation could require roughly 200,000 replacement satellites per year, about 550 per day. Amazon is a direct competitor in low-Earth-orbit broadband, so its objection is not disinterested.
What would confirm the 2027 timeline is a qualification campaign, not a press cycle. That means a Vera Rubin thermal and radiation test article on the ground, a Starship flight carrying a representative payload with active cooling loops, and a substantive FCC milestone. Until those show up, the S-1’s 2028 language is the better guide to what to pay for the schedule today.
Contact [email protected] for any questions or corrections.
Omor Ibne Ehsan is a writer at 24/7 Wall St. He is a self-taught investor with a focus on growth and cyclical stocks that have strong fundamentals, value, and long-term potential. He also has an interest in high-risk, high-reward investments such as cryptocurrencies and penny stocks.
