Skip to content
DopeSwagYolo

AI in Space

Nvidia and SpaceX Are Co-Designing Starmind's AI Computer, Huang and Musk Say

Jensen Huang said on a podcast released October 15, 2026 that Nvidia and SpaceX are designing a server for orbit and Earth. Elon Musk says the Star Mind computer reaches volume production in 2027.

By DopeSwagYolo4 min read

Researched and fact-checked by AI, with no human review. 13 sources listed below. How we verify

Nvidia and SpaceX engineers are designing a new server meant to work both in orbit and on the ground, Nvidia chief executive Jensen Huang said. He spoke on an episode of the BG2 podcast released on October 15, 2026. SpaceX chief executive Elon Musk replied on X the same day. He wrote that the "Star Mind" computer the two companies are co-engineering will reach volume production next year, TechNews reported.

The computer is meant for Starmind, SpaceX's planned network of AI computing satellites. Musk's latest public target for its first launch is the fourth quarter of 2027. SpaceX's request to deploy up to one million of the satellites was still under federal review as of September.

What did Huang and Musk say?

Huang said Nvidia and SpaceX will explore putting data centers in space together, Stocktwits reported. He said the two engineering teams are working on a server architecture that could suit both orbit and Earth. The Chinese financial news site Wallstreetcn reported the same remark. The podcast's own description lists whether SpaceX could help put data centers in orbit among its topics.

Huang also described a change in how Nvidia classes Musk's businesses, Stocktwits reported. Nvidia used to count them as enterprise customers, he said. It now groups them with its cloud service providers, the companies that rent computing power to others.

Musk wrote that the Star Mind computer is "far better than almost anyone expects." His post, as TechNews quoted it, gave no specifications, prices or launch dates.

What is Starmind supposed to be?

Here is what SpaceX and Nvidia have said about Starmind since August 2026:

  • August 4. On SpaceX's earnings call, Musk said each Starmind satellite would be essentially an optimized Vera Rubin NVL72 computer, Tom's Hardware reported. That is Nvidia's rack-scale system, which joins 36 Vera processors and 72 Rubin graphics processing units (GPUs). Musk said SpaceX also expects to use the design in ground data centers.
  • August 24. Nvidia said that SpaceXAI, as it calls SpaceX's AI business, will base its first-generation Starmind satellite on that system. It said orbit imposes very different limits on power, heat, bandwidth and reliability.
  • Late August. Musk posted that the system would launch to orbit in the fourth quarter of 2027, with significant scale in 2028, Light Reading reported. The outlet said the first AI1 model would be 20 meters high, with a 70-meter wingspan. It said the satellite would support peak power of 150 kilowatts.
  • September 18. SpaceX sent the U.S. Federal Communications Commission (FCC) a filing on re-entry and collision risk, SatNews reported. It put the mass of each Starmind node at up to 4,000 kilograms.
SpaceX plans Starmind satellites that would each carry an Nvidia rack-scale AI computer powered by large solar arrays.

Why does it matter?

The stated power is far above today's test hardware. Google's Project Suncatcher prototype, launched October 1, supplies about 1 kilowatt to its AI chips, CNN reported, citing The New York Times. Caleb Henry of the advisory firm Quilty Space said telecommunications satellites run on about 10 to 20 kilowatts, CNN reported. He said AI computing spacecraft may need as much as 100 kilowatts.

The two companies are also tied together financially. Nvidia disclosed in an August filing that it owned 122.8 million SpaceX Class A shares, CNBC reported. They were worth about $21 billion at the end of the second quarter. The shares came from Nvidia's investment in xAI, which SpaceX bought in February. Musk has said SpaceX will use only Nvidia chips in its AI data centers.

Orbital computing is also part of SpaceX's pitch to investors. Its IPO prospectus listed deploying orbital AI compute at scale among its growth strategies, Via Satellite reported. The filing warned that delays to the Starship rocket would hurt that strategy.

What is still unproven?

Heat is a central problem. A vacuum has no air to carry heat away, so a satellite must radiate it into space. At Nvidia's GTC conference in March, Huang said space offers no convection, only radiation. He said Nvidia had many engineers working on cooling, CNBC reported.

Regulatory approval is another open question. SpaceX applied on January 30, 2026 for up to one million satellites at altitudes of 500 to 2,000 kilometers. The FCC's Space Bureau accepted the application for filing on February 4 and asked for public comment. Acceptance is not approval. SpaceX also asked to be excused from the FCC's deployment milestone and surety bond rules. SatNews reported in September that the application remained under review.

Hardware also ages quickly. AI chips advance every one to two years, while satellites last five to seven years, according to a June report by JLL Research cited by CNN.

What comes next?

Both 2027 dates come from Musk: volume production of the Star Mind computer, and a first Starmind launch in the fourth quarter. SatNews described the first flights as prototype launches, subject to FCC licensing. SpaceX's own prospectus ties the wider plan to Starship.

The orbital data center tracker follows Starmind and rival projects, and labels flown hardware separately from plans. For how an orbital data center would work, and why critics doubt the economics, see the earlier explainer.

Sources

More from AI in Space

See all in AI in Space