Blab AI
SpaceX Unveils AI1: First Orbital Data Center with 72 Nvidia GPUs
SpaceX has turned the sky into a new frontier for computing, announcing the AI1 project – a satellite that will house 72 Nvidia GPUs and deliver 150 kW of processing power in orbit. The spacecraft, with a deployed wingspan of roughly 70 m and a height of about 20 m, is designed to become the world’s first orbital data center.
At its core, AI1 packs a formidable arsenal of 72 Nvidia Rubin GPUs, each equivalent to a GB300 server. Together they can reach a peak compute output of 150 kW, while the satellite is expected to run at an average load of 120 kW, giving a power density of roughly 70 kW per metric ton. To sustain this energy demand, the craft carries a 150 kW solar array that delivers about 250 W per square metre. Heat generated by the processors is shed through a deployable liquid‑cooled radiator that spans roughly 110 m², a design engineered to survive the vacuum of low‑Earth orbit.
AI1 is slated to cruise at an altitude of about 600 km, where it will operate within a radiation‑rich environment that presents significant engineering challenges. Because the Rubin GPUs are not yet radiation‑hardened, SpaceX may need to add shielding or accept a shorter mission life. Cooling in vacuum also adds mass and complexity, and the satellite’s 70‑metre wingspan and 110‑m² radiator contribute to a large, heavy payload.
SpaceX plans to begin production of the AI1 platform by the end of 2027, with the first launch targeted for 2028. The company is building a new manufacturing complex, dubbed Gigasat, in Bastrop, Texas. The 1,000‑acre site covers more than 11 million square feet and is intended to produce the AI satellites at scale.
The AI1 architecture is built around Nvidia’s Rubin Space‑1 module, unveiled at Nvidia’s GTC 2026 conference. The module is engineered for space‑grade operation and is said to deliver up to 25 times the AI compute of the H100 GPU. SpaceX’s own statement notes that the first AI1 units will use Rubin GPUs, equivalent to a GB300 server, in a configuration that supports 72 GPUs on a single spacecraft.
SpaceX frames the AI1 platform as a response to the escalating energy demands of terrestrial data centers. The company’s spokesperson said that moving compute to orbit could relieve pressure on Earth‑based power grids, although the satellite will still rely on solar power and will need to manage energy storage for eclipse periods.
Potential customers for an orbital data center include SpaceX’s own AI ventures, such as xAI, as well as defense agencies that value the isolation of space‑based compute from ground‑based attacks. Commercial cloud providers—Microsoft, Google, Amazon—could also be interested if orbital compute proves cost‑competitive. Niche applications that benefit from proximity to space assets, such as real‑time satellite imagery processing, climate monitoring, and large‑constellation orchestration, are also mentioned.
Several technical hurdles remain. The high‑energy radiation environment of low‑Earth orbit can degrade electronic components, and the Rubin GPUs are not yet radiation‑hardened. SpaceX may need to add shielding or accept shorter satellite lifetimes. Cooling in vacuum is another hurdle; the large radiator surface required for 150 kW of heat dissipation adds mass and complexity. Latency between orbit and Earth is higher than for terrestrial data centers, limiting suitability for low‑latency workloads.
Launch cost is also a factor. While SpaceX’s reusable rockets have lowered launch expenses, deploying a 70‑meter satellite still represents a significant outlay. The company has not yet disclosed a detailed cost model for orbital compute versus ground‑based alternatives.
Regulatory and jurisdictional questions are unresolved. It is unclear under which national or international laws an orbital data center would operate, and how data processed in space would be governed.
SpaceX’s AI1 project marks a shift from using space primarily for transport and communication to using it as an active computational platform. If the project succeeds, it could mark the beginning of a new industry that places AI workloads beyond Earth’s atmosphere.
As of now, the AI1 project remains in an early stage, with no firm launch date beyond the 2028 target. The company’s Gigasat factory is under construction, and the Rubin module has been announced by Nvidia. The next steps will involve validating the space‑grade performance of the GPUs, demonstrating the cooling system, and proving the satellite’s ability to operate in the radiation environment of low‑Earth orbit.
The outcome of these efforts will determine whether orbital data centers become a viable complement to terrestrial data centers, or remain a speculative concept.
At its core, AI1 packs a formidable arsenal of 72 Nvidia Rubin GPUs, each equivalent to a GB300 server. Together they can reach a peak compute output of 150 kW, while the satellite is expected to run at an average load of 120 kW, giving a power density of roughly 70 kW per metric ton. To sustain this energy demand, the craft carries a 150 kW solar array that delivers about 250 W per square metre. Heat generated by the processors is shed through a deployable liquid‑cooled radiator that spans roughly 110 m², a design engineered to survive the vacuum of low‑Earth orbit.
AI1 is slated to cruise at an altitude of about 600 km, where it will operate within a radiation‑rich environment that presents significant engineering challenges. Because the Rubin GPUs are not yet radiation‑hardened, SpaceX may need to add shielding or accept a shorter mission life. Cooling in vacuum also adds mass and complexity, and the satellite’s 70‑metre wingspan and 110‑m² radiator contribute to a large, heavy payload.
SpaceX plans to begin production of the AI1 platform by the end of 2027, with the first launch targeted for 2028. The company is building a new manufacturing complex, dubbed Gigasat, in Bastrop, Texas. The 1,000‑acre site covers more than 11 million square feet and is intended to produce the AI satellites at scale.
The AI1 architecture is built around Nvidia’s Rubin Space‑1 module, unveiled at Nvidia’s GTC 2026 conference. The module is engineered for space‑grade operation and is said to deliver up to 25 times the AI compute of the H100 GPU. SpaceX’s own statement notes that the first AI1 units will use Rubin GPUs, equivalent to a GB300 server, in a configuration that supports 72 GPUs on a single spacecraft.
SpaceX frames the AI1 platform as a response to the escalating energy demands of terrestrial data centers. The company’s spokesperson said that moving compute to orbit could relieve pressure on Earth‑based power grids, although the satellite will still rely on solar power and will need to manage energy storage for eclipse periods.
Potential customers for an orbital data center include SpaceX’s own AI ventures, such as xAI, as well as defense agencies that value the isolation of space‑based compute from ground‑based attacks. Commercial cloud providers—Microsoft, Google, Amazon—could also be interested if orbital compute proves cost‑competitive. Niche applications that benefit from proximity to space assets, such as real‑time satellite imagery processing, climate monitoring, and large‑constellation orchestration, are also mentioned.
Several technical hurdles remain. The high‑energy radiation environment of low‑Earth orbit can degrade electronic components, and the Rubin GPUs are not yet radiation‑hardened. SpaceX may need to add shielding or accept shorter satellite lifetimes. Cooling in vacuum is another hurdle; the large radiator surface required for 150 kW of heat dissipation adds mass and complexity. Latency between orbit and Earth is higher than for terrestrial data centers, limiting suitability for low‑latency workloads.
Launch cost is also a factor. While SpaceX’s reusable rockets have lowered launch expenses, deploying a 70‑meter satellite still represents a significant outlay. The company has not yet disclosed a detailed cost model for orbital compute versus ground‑based alternatives.
Regulatory and jurisdictional questions are unresolved. It is unclear under which national or international laws an orbital data center would operate, and how data processed in space would be governed.
SpaceX’s AI1 project marks a shift from using space primarily for transport and communication to using it as an active computational platform. If the project succeeds, it could mark the beginning of a new industry that places AI workloads beyond Earth’s atmosphere.
As of now, the AI1 project remains in an early stage, with no firm launch date beyond the 2028 target. The company’s Gigasat factory is under construction, and the Rubin module has been announced by Nvidia. The next steps will involve validating the space‑grade performance of the GPUs, demonstrating the cooling system, and proving the satellite’s ability to operate in the radiation environment of low‑Earth orbit.
The outcome of these efforts will determine whether orbital data centers become a viable complement to terrestrial data centers, or remain a speculative concept.
