The AI race has its sights on the stars, as companies push to migrate the world's networks to orbit. It's sold as a catchall solution for many of the industry's limiting factors, including land scarcity and environmental concerns and power shortages.

One of the movement's most vocal proponents is Elon Musk, whose firm SpaceX recently filed an application with the FCC to launch a data center with up to a million satellites into orbit. Dubbed Starmind, SpaceX's orbital data center project looks to be the company's most ambitious yet.

Eventually, Starmind will likely depend on SpaceX's Terafab, a $119 billion joint chip manufacturing plant with Musk's Tesla. In the meantime, however, SpaceX will lean on NVIDIA to meet its expedient timeline. During the earnings call, Musk said that the Starmind AI1 satellite "is not some sort of far future distant thing," claiming that SpaceX will begin launching in 2027 — an optimistic timeline, even for the notoriously optimistic Musk. 

Despite the executive's confidence, critics remain skeptical. One issue likely to challenge the feasibility of the project is the decision to opt for NVIDIA's most-advanced chip suite, as technical hurdles, cost constraints and scaling issues threaten to push the constellation past next year's launch date.

Before diving into the difficulties of deploying the latest chip architecture in space, let's quickly contextualize this decision within SpaceX's orbital data center ambitions. Starmind's AI1 satellite will look to beam AI results back to Earth via "high-speed laser links" that utilize SpaceX's Starlink system. Each satellite is expected to hold roughly 72 NVIDIA chips, making one unit the equivalent of a single NVIDIA server rack. SpaceX predicts each will produce up to 175kW of computing power on average. Ultimately, SpaceX aims to deploy to a million of these massive satellites in Low Earth Orbit.

During the company's Q2 earnings call in early August, Musk announced that NVIDIA will provide the chip architecture behind its AI1 project, the initial salvo of satellites in SpaceX's Starmind fleet. According to a SpaceX post on X, AI1 satellites will use "NVIDIA Rubin GPUs and Vera CPUs for data center class space compute." SpaceX also stated that it will use NVIDIA's Vera Rubin NVL72 rackscale system as the project's chip platform. Dubbed Space-1, NVIDIA claims that the Vera Rubin module delivers up to 25 times more AI compute than its older H100 GPU model.

SpaceX hopes the constellation will untether xAI's growth from "Earth's grid, land, and cooling constraints." Following a sun-synchronous orbit that keeps it facing the sun 98 percent of the time, AI1's massive solar panel array will reportedly to generate 210kW of power. In doing so, the company will solve the power issue plaguing terrestrial data centers. The satellites will generate massive amounts of heat, an issue SpaceX plans to solve with 1,700-square-foot liquid radiators that release heat "freely into the vacuum of space." Whether SpaceX can do so more efficiently than current space infrastructure projects remains a major challenge. 

During the recent earnings call, Musk detailed the logic behind SpaceX's decision to bank its Starmind project on NVIDIA's Vera Rubin chips, calling it the "best architecture" available. "We think the design of the NVL72 VR computer is a much better design than, say, having a standard rack-style design," Musk continued, noting that the company expects to "deploy this on the ground as well as in orbit." During the call, the executive claimed that the Vera Rubin design would also "cost less" and "be more effective" than other solutions.

But putting the most advanced AI chips in orbit might not render the best results. Speaking to Engadget in February, Dr. Benjamin Lee, a professor at University of Pennsylvania's Department of Electrical and Systems Engineering, warned that using advanced chip architecture could expose satellites to unnecessary data corruption. 

Concerns surrounding newer chips stems from the fact that they are more susceptible to a phenomenon known as bit flips in which the radiation from solar weather and other common events low earth orbit force the ones and zeros that make up the chip's binary code to swap values. This susceptibility derives from new chips' use of smaller transistors, a trend that has enabled engineers to reach new heights of computational power. As Dr. Lee explained, "ones and zeros are really about the presence or absence of electrons and the amount of charge required to represent a one goes down as the transistors get smaller." The downsizing of transistors has transformed a comparatively minor issue into a widespread problem for semiconductors operating in our atmosphere.

Proponents of orbital data centers, in turn, argue that companies like SpaceX have likely built contingency plans for withstanding radiation. For one thing, AI systems are built to detect and correct for these data corruption errors. Starcloud, a startup that launched NVIDIA H100 GPUs into orbit November last year, believes that orbital data centers could have the same shelf life as terrestrial ones. Google, for its part, found that its V6e Trillium TPU compute tray could likely operate reliably in orbit for five years.

But Lee notes that these arguments somewhat miss the mark. According to the professor, the question isn't only whether Starmind will be able to detect and correct bit flip errors, but rather, if doing so will make orbital data centers comparatively inefficient. As Dr. Lee pointed out in February, "even though modern computer architectures can detect and sometimes correct for those errors, having to do that again and again will slow down or add an overhead to space-based computation." Ultimately, it is less likely that radiation dooms projects like Starmind than it will add inefficiencies inherently absent from its terrestrial rivals.

For his part, Musk has downplayed the proposed difficulty of getting the AI1 project off the ground. In a promotional video in June, Musk said that "there's not some magic that's necessary that doesn't exist" to get Starmind into orbit. Instead, the executive argued "a lot of this is technology we've already made for Starlink V3 satellites." 

But several hurdles remain. The most obvious roadblock is financial. According to a July 2026 policy brief by management consulting firm Bain, data center launches need to cost somewhere between $50 and $100 per kilogram to be economical. As it stands, SpaceX's Falcon Heavy rocket can only launch payloads at a cost ratio of roughly $1,500 per kilogram, a far cry from profitability. Further complicating matters is that the AI1 satellite is too big to launch with SpaceX's Falcon 9, placing all its hopes on the work-in-progress Starship rocket. Moreover, SpaceX will likely need to scale its daily launch rates far beyond the current global total to maintain its constellation. To make Musk's data center dreams a reality, SpaceX will need to launch larger satellites more efficiently at previously unfathomable rates.

Once in orbit, Starmind is likely to run into problems beyond fending off radiation-fueled data corruption. For one thing, maintenance is likely to be difficult, with many believing SpaceX will launch new satellites with upgraded technologies rather than repair or update old ones. A study Meta published following the release of its Llama 3 model detailed hardware-related failures happening as often as every three hours during training, and that was on earth. For space-based data centers this could pose an acute problem that becomes paramount as low earth orbit becomes more crowded. With a proposed 1 million satellites potentially entering low earth orbit from SpaceX alone, some experts are worried about the likelihood of collisions with space debris and other satellites. Others, meanwhile, bemoan the adverse environmental consequences such a cluttered sky may bring.