NVIDIA và SpaceX hợp tác phát triển vệ tinh Starmind AI1 cho các trung tâm dữ liệu AI quỹ đạo

The Huoxing Finance report states that on August 5, NVIDIA and SpaceX are collaborating to develop an orbital computing satellite named Starmind AI1, aiming to deploy an AI data center in space equipped with NVIDIA’s next-generation Rubin GPU and Vera CPU to deliver data center-grade computing power in the space environment. SpaceX indicates that the Starmind project seeks to build an orbital AI computing network powered by solar energy and connected to Earth via high-speed laser links between Starlink satellites, realizing an “orbital data center” architecture. The project aims to migrate some AI computation from Earth to orbit to alleviate current limitations in land use, power supply, and heat dissipation faced by AI infrastructure. Disclosed details reveal that when deployed, Starmind AI1 will stand approximately 30 meters tall with a wingspan of about 75 meters, reaching a peak computing power consumption of 250 kilowatts and an average of 175 kilowatts, with its overall design resembling more of a “flying computing facility.” However, orbital AI data centers still face numerous challenges, including radiation exposure, hardware reliability, communication latency, equipment maintenance, and launch costs. To date, no company has demonstrated that large-scale orbital data centers can outperform ground-based data centers in terms of economic viability and operational efficiency. NVIDIA’s involvement in this initiative further expands its AI infrastructure footprint. The Rubin GPU and Vera CPU were originally designed for next-generation AI data centers, and the Starmind project will attempt to deploy these high-performance computing chips into the orbital environment. SpaceX’s advantage lies in its proprietary rockets and satellite network, which can reduce launch costs through Falcon 9 and future Starship vehicles, while integrating computing, communication, and launch capabilities via Starlink. The report notes that SpaceX has previously proposed building up to one million orbital data center satellites in the future. If this plan progresses, it could reshape the development trajectory of AI infrastructure but may also trigger issues such as orbital congestion, space debris, and regulatory concerns. Analysts believe that in the short term, Starmind is more likely to be applied in scenarios such as satellite imagery processing, scientific computing, and defense. Whether it can evolve into a true “orbital cloud computing network” will depend on cost, reliability, and scalability.