Starmind is a planned constellation of satellites designed to run artificial intelligence workloads in orbit, announced by SpaceX on August 4, 2026.[1][2] The first satellite design, Starmind AI1, is being developed with NVIDIA around the chipmaker's Rubin GPUs and Vera CPUs, and SpaceX says launches will begin in 2027.[1][3] A January 2026 application to the US Federal Communications Commission seeks authority for up to one million of the satellites, far more spacecraft than any constellation flown to date.[4] No Starmind hardware has reached orbit, and every published figure is a design target rather than a demonstrated capability.
The program is the orbital arm of SpaceX's expansion into AI infrastructure, which followed its acquisition of the AI company xAI in February 2026.[5] It is the largest of several orbital data center projects announced since 2025; NVIDIA also supplies the startup Starcloud and other companies building computing hardware for space.[6]
Announcement
SpaceX revealed the program in a post on X on August 4, 2026, saying it was partnering with NVIDIA to design the AI1 compute payload with "Nvidia Rubin GPUs and Vera CPUs for datacenter class space compute."[1][2] The post landed the same day as SpaceX's first earnings report as a publicly traded company, covering the second quarter of 2026.[2][3]
On the earnings call, Elon Musk committed SpaceX's AI buildout to a single chip supplier: "We're exclusive to Nvidia," he said, adding "We think the Vera Rubin architecture is the best AI computer."[3] He told investors the company expects more than 2 gigawatts of installed compute, most of it on the ground, by the end of 2026, rising to roughly 10 gigawatts, and perhaps 20, by the end of 2027. Those figures are company projections, not capacity that exists today.[3][7]
The corporate backdrop is unusual for a satellite program. SpaceX announced its acquisition of xAI on February 2, 2026, an all-stock deal that valued the combined company at about 1.25 trillion dollars and folded the Grok chatbot and the X platform into SpaceX.[5] The AI unit completed its rebrand to SpaceXAI in July 2026.[8]
The AI1 satellite
SpaceX says a deployed AI1 will stand about 30 meters tall, span about 75 meters from solar array tip to tip, and support a compute payload of roughly 250 kW.[2][9] That is approximately the power drawn by one of NVIDIA's Vera Rubin NVL72 racks, and Musk has described the satellite as "essentially an optimized Vera Rubin NVL72 computer."[2][7]
The payload uses NVIDIA's Space-1, a version of the Vera Rubin platform adapted for orbit.[2] NVIDIA introduced its space computing line in March 2026, claiming the Rubin GPU in the Space-1 Vera Rubin module delivers up to 25 times the AI compute of its H100 for space-based inference.[6]
Musk has said the first Starmind satellites will launch in 2027.[3][7] Two prototype satellites are planned for early 2027, with mass production intended to follow later that year.[10] SpaceX has not published a more detailed schedule.
FCC application
The regulatory groundwork predates the Starmind name. In January 2026, SpaceX applied to the FCC, under its licensing name Space Exploration Holdings, for a "SpaceX Orbital Data Center System" of up to one million satellites, orbiting between 500 and 2,000 km at 30-degree and sun-synchronous inclinations in shells up to 50 km deep, with Ka-band links for ground communications.[4][11] The application does not use a satellite brand name.[11] It argues the constellation would achieve "transformative cost and energy efficiency" by "directly harnessing near-constant solar power with little operating or maintenance cost," and it asks the commission to waive the standard deployment milestones that normally require half of a constellation to fly within six years.[11]
The FCC's Space Bureau accepted the application for filing on February 4, 2026, a procedural step that opened it to public comment with a March 6 deadline.[4] The commission has not authorized the system.
Starlink and Starship
The application leans on Starlink. SpaceX cites its experience operating that constellation, and the compute satellites would use optical inter-satellite links to communicate with each other and with Starlink spacecraft, which would relay results to the ground.[11]
Launch cost is the other dependency. At Falcon 9 prices of about 7,000 dollars per kilogram, The Register estimated a single AI1 launch at more than 23 million dollars; at the 10 dollars per kilogram Musk targets for Starship, the same launch would cost around 33,000 dollars.[2] The FCC application likewise emphasizes Starship's deployment capacity, so the constellation's economics rest on a launch price no rocket has yet approached.[11]
Reception and concerns
Not everyone believes computing in orbit can compete with terrestrial data centers on cost: a Gartner analyst quoted by The Register called space-based data centers "peak insanity."[2] The announcement also came with financial strain attached: SpaceX reported a 541 million dollar net loss on 7.81 billion dollars of quarterly revenue, with capital spending of 18.4 billion dollars, about six times the year-earlier figure, driven by the AI buildout.[2][9]
Astronomers objected to the million-satellite application within weeks of its filing. Dark-sky consultant John Barentine and a group of colleagues filed a challenge with the FCC before the March 6 comment deadline, warning that the proposed orbits are high enough to stay sunlit even at midnight and could put tens of thousands of star-bright moving objects in the night sky at once; with a million satellites overhead, he argued, the shutter at a major observatory "would be closed more than it would be open."[12] The European Southern Observatory calculated that the Very Large Telescope could lose up to 10 percent of pixels in every image, and up to 30 percent for some observations, while the Royal Astronomical Society's Robert Massey called the plan "really intolerable."[13] Tony Tyson, chief scientist of the Vera C. Rubin Observatory, warned in June 2026 of "major impacts on optical astronomy from 2027 on."[14] Critics also note that maintaining a fleet that size would mean a retired spacecraft reentering roughly every three minutes, adding aluminum oxide to the upper atmosphere and adding collision and space debris risk in already crowded orbits.[12]
References
- SpaceX announcement of the Starmind AI1 partnership with NVIDIA - SpaceX on X, August 4, 2026.
- Elon pledges to give Nvidia a virtual monopoly over the stars - The Register, August 5, 2026.
- Nvidia stock rises after Musk says SpaceX will exclusively use company's chips - Yahoo Finance, August 5, 2026.
- Space Bureau Accepts for Filing SpaceX's Application for Orbital Data Centers (DA-26-113) - Federal Communications Commission, February 4, 2026.
- Elon Musk's SpaceX acquiring AI startup xAI ahead of potential IPO - CNBC, February 2, 2026.
- NVIDIA Launches Space Computing, Rocketing AI Into Orbit - NVIDIA Newsroom, March 16, 2026.
- SpaceX earnings: 2GW of compute by year end, aiming for 10-20GW by end of 2027 - Data Center Dynamics, August 2026.
- xAI is now officially known as SpaceXAI - Engadget, July 7, 2026.
- SpaceX stock hits new all-time low as AI capex jumps in Q2 - Yahoo Finance, August 6, 2026.
- SpaceX (SPCX) Q2 2026 earnings: Soaring AI costs outweigh revenue beat in first report since IPO - CNBC, August 4, 2026.
- SpaceX files plans for million-satellite orbital data center constellation - SpaceNews, January 2026.
- SpaceX plan for 1 million orbiting AI data centers could ruin astronomy, scientists say - Space.com, March 13, 2026.
- "This is really intolerable": Astronomers protest giant orbiting mirror project and SpaceX's million AI satellites - Space.com, March 22, 2026.
- Astronomers fear orbital data centers will interfere with observations - SpaceNews, June 12, 2026.
