SpaceX and Tesla CEO Elon Musk said on the 24th that he will launch the first "AI satellite" powered by Nvidia's latest artificial intelligence accelerator into space in the fourth quarter of 2027.
Musk said in a post on X that SpaceX and Nvidia have designed a Vera Rubin NVL72 system optimized for the space environment. According to the prototype design Musk released, a single AI satellite delivers up to 150 kilowatts of computing power, equivalent to one rack of Nvidia's Blackwell (GB300) system used in ground data centers. The energy needed to run the computing resources is to come from solar power, and the satellite's wingspan with solar panels attached reaches 70 meters.
SpaceX has laid out a roadmap to build 1 gigawatt of AI computing infrastructure in space annually by the end of 2027, then expand tenfold each year to eventually reach terawatt scale. The company has applied to the Federal Communications Commission for approval to launch as many as 1 million satellites for space data centers.
But critics point out that the goal and reality are two different things. Building a data center in space is a problem of an entirely different order from installing server racks in a ground facility. That is precisely why Nvidia's Vera Rubin NVL72 had to be redesigned for space.
The tech industry expects the design to sharply strengthen durability against radiation and other external factors. Solar wind and cosmic radiation pour down unfiltered on the orbit where the satellites will fly. When high-energy particles strike a semiconductor's silicon chip, the phase of electrons trapped in transistors flips. That causes data errors that are fatal to the complex calculations AI performs, and can go further to permanently damage the chip's physical circuitry.
Thermal management also needs improvement. Space is cold, but cooling an object there is harder than on the ground. On Earth, technology for cooling heated servers continues to improve, whether by air or liquid. In the vacuum of space, however, convection cooling is fundamentally impossible. That makes radiating heat away critical. SpaceX has previously said it can solve the heat problem by applying enormous radiator panels when it deploys AI satellites.
Weight is another variable. Loading cargo onto a rocket costs tens of millions of won per kilogram. Musk stressed that the Vera Rubin-based rack for space is "much simpler, lower cost, higher density and lighter than existing racks." That means custom lightweighting technology has been applied, stripping away unnecessary casings and pushing integration density to the limit for launch.
Skepticism remains widespread. Market research firm Wood Mackenzie estimates that building a 1-gigawatt space data center would require $170 billion (about 226 trillion won), more than three times the cost of a ground data center. Analysts say enormous launch costs and the steep price of space-grade semiconductors will hold the project back.
The industry is watching whether Musk can cut costs with reusable launch vehicle technology. On that point, Musk has declared: "I actually think that the cost of deploying AI in space will drop below the cost of terrestrial AI much sooner than most people expect. I think it may be only two or three years."
