SpaceX and Nvidia Plan to Send AI Data Centers to Orbit by 2027

SpaceX, in partnership with Nvidia, is preparing to send a fully operational AI data center into low Earth orbit, marking a significant step in the convergence of space exploration and high-performance computing. Elon Musk revealed on X that a "space-optimized Vera Rubin NVL72 system" designed by SpaceX and Nvidia is slated for launch in the fourth quarter of next year, with large-scale deployments following in 2028.

The NVL72 is a dense rack-scale system containing 72 Rubin GPUs, 36 Vera CPUs, and advanced networking components including ConnectX-9 SuperNICs and BlueField-4 DPUs. Each unit weighs approximately 1,800 kilograms (4,000 pounds)—roughly the weight of a pickup truck—and draws 120 kilowatts of power. Nvidia notes the system uses novel liquid cooling, while SpaceX has published designs for a dedicated data center satellite equipped with a 210-kilowatt solar array and liquid radiators designed to reject heat into the vacuum of space.

SpaceX’s Falcon 9 rocket, capable of lifting 22,800 kilograms to low Earth orbit, could accommodate several NVL72 racks per launch, given the rack dimensions of roughly 600 mm by 2,293 mm by 1,068 mm. With SpaceX’s demonstrated ability to launch over 100 Falcon 9 missions annually, the logistical capacity appears feasible. However, neither company has specified which rocket will be used for the initial payload or the exact modifications required to make the NVL72 "space-optimized."

The Register notes that the term likely encompasses radiation hardening and structural reinforcement to withstand launch stresses, but further details were not provided. Critics point to Musk's history of overpromising: The New York Times recently found that Musk delivered on only 19 percent of 602 public promises and was either years late or never fulfilled 35 percent of them. While SpaceX has a strong track record of innovation in rocketry and satellite deployment, building and operating a functional data center in space presents unprecedented technical challenges, including thermal management, maintenance, and reliable power generation in orbit.

Why This Matters

This could enable real-time AI processing in orbit, reducing latency for applications like Earth observation, autonomous navigation, and communications.
A successful deployment would make space-based cloud computing a reality, potentially disrupting terrestrial data center economics and enabling new classes of space applications.
The project tests whether cutting-edge terrestrial hardware can be ruggedized for space at scale, with implications for future lunar and Martian infrastructure.

Background

The concept of space-based data centers has been explored for decades, but high launch costs and hardware limitations made it impractical. The recent proliferation of low Earth orbit satellite constellations—like SpaceX's Starlink—has reduced launch costs and proven orbital infrastructure at scale. Meanwhile, the AI boom has created immense demand for compute power, driving companies to seek unconventional locations for data centers, including underwater and near nuclear plants. Nvidia's Vera Rubin architecture, announced as a successor to Blackwell, is designed for high-density AI workloads. SpaceX has experience operating complex vehicles on orbit, including the Dragon spacecraft and Starship prototypes, but has not previously attempted to host third-party compute racks.

Key Perspectives

SpaceX and Nvidia: They frame the project as a natural extension of their respective capabilities—SpaceX in orbital logistics and Nvidia in AI hardware—and a solution to the growing need for orbital compute. The solar array and liquid cooling designs suggest they have solved key engineering hurdles. Aerospace and data center engineers: The extreme environment of space—radiation, vacuum, thermal cycling, and vibration during launch—poses severe challenges. Liquid cooling in microgravity is untested at this scale, and in-orbit maintenance is nearly impossible for complex electronics. Critics and analysts: Musk’s track record of missed deadlines and overpromises raises skepticism. Experts also question the economic viability, as terrestrial data centers offer far easier access for upgrades and repairs. Space-based compute may be cost-effective only for niche applications that require low latency from orbit.

What to Watch

Whether SpaceX or Nvidia release concrete specifications on radiation hardening, power management, and thermal control for the space-optimized NVL72.
The choice of launch vehicle (Falcon 9 vs. Starship) and whether the first launch occurs by Q4 2027 as promised.
Demonstration of in-orbit computing capabilities, such as real-time image processing or AI inference, from the first deployed unit.
Any regulatory or spectrum allocation issues related to operating high-power computing hardware in space.

ZOTPAPER