Musk announced that the space-based Vera Rubin NVL72 system, developed by SpaceX and NVIDIA, will be launched into orbit in Q4 2027 and enter large-scale deployment in 2028; however, the real challenge for this system to deliver its target computational power lies in managing massive heat dissipation in a vacuum environment.
(Prior context: SpaceX exclusively locks in NVIDIA's AI architecture; space data center Starmind project to launch next year)
(Background supplement: How does SpaceX’s space data center cool itself? Engineers scoffed that it’s impossible; Musk: Over ten thousand satellites are already operational)
Elon Musk announced this morning (25th) on X that SpaceX, in collaboration with NVIDIA, has designed a space-specialized Vera Rubin NVL72 system, set to launch into orbit in Q4 2027 and enter large-scale deployment in 2028.
NVIDIA stated that SpaceX's deployment of the Vera Rubin NVL72 packs 72 Rubin GPUs and 36 Vera CPUs into a single rack, interconnected with sixth-generation NVLink, achieving a total intra-rack bandwidth of 260 TB/s.
The NVL72 was originally designed by NVIDIA as a flagship rack for ground-based AI factories, and this is its first transformation into a space-grade specification—essentially deploying the most advanced data center configuration directly on a satellite. NVIDIA also announced that companies under Musk’s umbrella will adopt the Vera CPU to accelerate computing for Grok and the next generation of “agent AI.”
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
Reasons to bet on NVIDIA
The development of this system underwent a clear turning point from conception to finalization. In early August, SpaceX announced it would exclusively adopt NVIDIA technology; prior to that, Musk had stated he would seek different chips and design solutions, comparing options.
The final decision was to bet everything on NVIDIA for the entire space data center. The logic behind this decision may not be complicated: the space environment does not allow for trial and error; rack volume, weight, and thermal pathways must all be redesigned, and supporting multiple architectures would multiply development time several times over, requiring repeated validation of the reliability of multiple systems.
The cost logic of space launches is also straightforward: every additional kilogram of weight means higher launch costs, and maintaining multiple chip solutions means paying separately for different weight and thermal designs. In his post, Musk described this satellite-based data center as "significantly simpler, lower cost, higher density, and lighter" than traditional rack-based data centers.
Heat dissipation is the real challenge.
The problem is that there is no air in space. Ground-based data centers rely on air convection—using fans and cooling towers—to dissipate heat; but in a vacuum, heat can only be removed by liquid cooling loops that transfer the heat to radiators, which then emit the thermal energy as infrared radiation into deep space—an approach that has been used on the International Space Station for decades.
SpaceX's Starmind AI1 prototype satellite achieves a peak computing power of 150 kW using a single 110-square-meter deployable liquid-cooled heat dissipation panel, with a wingspan of 70 meters—longer than that of a Boeing 747—orbiting at approximately 600 kilometers.
The questioner did the math: dissipating just 1 MW of waste heat requires about 1,200 square meters of radiators—equivalent to four tennis courts. If we scale this to a 100 MW AI data center, the minimum viable ground-level size, the radiator area would surge to 100,000 square meters. In contrast, SpaceX’s own targets call for 2 GW of computing power by the end of 2026 and nearly 10 GW by the end of 2027—the gap here is orders of magnitude, not something that can be bridged by adding a few more radiator panels.
In response to concerns about heat dissipation, Musk did not directly address how the issue would be resolved; instead, he pointed out that SpaceX’s in-orbit satellites have exceeded 10,000, accounting for about two-thirds of all active satellites worldwide—a scale far exceeding the combined total of all other space systems. This statement demonstrates launch and deployment capabilities, not heat dissipation capabilities—two entirely different issues.
In the end, the space data center competition isn't about who launches a rack into space first, but rather who achieves the lowest launch cost, largest cooling surface area, and highest compute power per watt—when will it become cheaper than on Earth? That depends on whether Musk can work another miracle.
📍Related Articles📍
SpaceX has officially renamed itself SpaceXAI, with Musk tying space and AI together as one.