Data Centers Look to Space for Growth
As terrestrial data centers face growing power and space constraints, some are looking to orbit as the next frontier. The Economist reports that putting constellations of data centers in space is now being considered a viable idea to leverage solar power and overcome land-based limits, a potential long-term scaling play for AI and govtech infrastructure.
The explosive growth of AI is a primary driver for moving data centers to space, with AI workloads consuming up to 1,000 times more electricity than traditional web searches. Terrestrial data centers are projected to double their electricity consumption by 2030, putting immense strain on local power grids and water resources. A single large AI model's training can use the energy equivalent of hundreds of households over several months. Startups are already launching proofs of concept. Starcloud, a member of NVIDIA's Inception program, launched a satellite in November 2025 with an NVIDIA H100 GPU to test its viability for running AI models in orbit. The company successfully trained a NanoGPT model in orbit and has demonstrated that AI queries can be performed in space. Following this, Google's Project Suncatcher is set to launch two satellites in early 2027 to test its own proprietary TPUs, designed for large AI workloads, in the high-radiation environment of low-Earth orbit. The advantages of orbital data centers extend beyond overcoming terrestrial energy constraints. They offer access to continuous and more intense solar power, natural cooling in the vacuum of space, and enhanced security from terrestrial threats. Furthermore, placing data processing in orbit can reduce latency for satellite-based applications and eliminate the bottleneck of downlinking massive amounts of raw data from Earth observation satellites. However, significant challenges remain. The vacuum of space makes heat dissipation difficult, requiring large radiators, and radiation exposure can degrade hardware. Maintenance and hardware upgrades are currently impractical, and launch costs, despite reductions by companies like SpaceX, are still substantial. Several companies are tackling these challenges head-on. SpaceX is exploring the use of its Starlink satellites for data processing and is seeking funding to build out this capacity. Lonestar Data is pursuing an even more ambitious goal of establishing data centers on the moon for secure data backup. Meanwhile, Aetherflux, founded by Robinhood co-founder Baiju Bhatt, is developing an orbital data center satellite constellation called "Galactic Brain" and aims to launch its first commercial node in 2027.
