Co-written with Alessandro Secchi
The idea of placing AI data centers in space is slowly moving from science fiction to reality. In late 2025, Elon Musk’s SpaceX and Google announced plans to launch into orbit a server that will train and run LLMS. This science fiction-like vision is driven by legitimate concerns that AI data centers require more energy and land than is available.
According to technology and space leaders, the benefit of building an AI data center in space is that it addresses sustainability and capacity issues raised by the laws of scale in LLM training. For example, GW data centers, the industry’s next major milestone, will consume approximately 8.8 to 10 TWh per year, including cooling and other overheads. Space-based data solutions offer an alternative solution to harnessing solar energy directly in orbit. Additionally, relocating AI infrastructure into space could help companies address size and scale limitations. Data centers on Earth are limited by land availability, local infrastructure, and the risk of exacerbating local water scarcity and raising utility costs for surrounding communities. In orbit, technology and space leaders claim that the “real estate” is virtually limitless. The move to space could also create a “clean slate” for governance, allowing for data sovereignty and new rules for operating outside traditional jurisdictions.
The move to space orbit also takes advantage of recent advances in inference scaling. New data shows that scaling inference computations can be more efficient than increasing model size. Inference workloads are easily distributed as long as inference is performed locally within a tightly connected cluster. This is where companies launching orbital infrastructure could potentially benefit. By integrating GPUs into existing satellite constellations, you can create a large number of distributed computing nodes in orbit. For example, SpaceX may evolve its communications network into a distributed orbital cloud.
While it’s heartening to see technology leaders thinking outside the box to address AI’s energy demands and sustainability challenges, there are big questions about the idea’s feasibility. Experts interviewed by The New York Times say launching hardware into orbit remains prohibitively expensive. Currently, SpaceX’s Falcon 9 launch cost exceeds $100 billion (based on $1,400/kg). Similar projects, such as SpaceX’s Starship project, need to achieve launch costs of $200/kg. This is often cited as the threshold at which orbital computing can compete economically with data centers on Earth. Additionally, modern chips are not designed to withstand cosmic radiation and require large radiator systems to cool them in a vacuum, as there is no air to dissipate the heat. These orbital data centers also need to be replaced every five years due to the lifetime of the hardware. While progress has been made on these fronts, such as Google’s Project Suncatcher demonstrating that its Trillium TPU can withstand low-Earth orbit radiation over a five-year mission life, major engineering and economic hurdles remain.
Potential impact on industry
For now, former SpaceX executive Tom Mueller said in an interview with The New York Times that the idea could focus on a long-term vision and capitalize on the buzz around both AI and space. However, if space-based AI data centers become a reality in the future, the ripple effects could reach a variety of sectors.
If tech companies achieve more efficient AI computing in space, they could be in a position to offer their services at lower costs or with greener credentials than operators on Earth. Cloud service and data center operators are likely to prepare for hybrid models that incorporate some workloads in space in the moderate future. This includes planning workload orchestration across the department and establishing an innovation team to track space computing developments and potentially pilot small-scale experiments.
For satellite carriers and communications providers, space-based data centers can be partners or competitors. An important question to consider is how orbital computing can help enhance communication networks and diversify their offerings. Those working in the semiconductor industry may be aware of the need for more radiation-hardened chips that can operate in space. Those involved in thermal components can also look forward to new markets for liquid cooling systems designed for space.
For most business leaders relying on AI rather than building infrastructure, a key question is how space-based data centers will impact access to computing power, cost structures, and sustainability goals. In the short term, little will change. However, as with any emerging technology, it is wise to maintain strong relationships with vendors and seek out pilot opportunities to gain first-hand experience. More broadly, this trend is driving forward-looking thinking about digital supply chains and how companies can procure and scale AI services in a fundamentally different computing environment.
While still speculative, the push to move AI infrastructure off Earth reflects how urgent and complex the challenge of scaling AI has become. Business leaders don’t need to bet on trajectory now, but they must remain adaptable, embrace new partnerships, and be ready to rethink where their computing power comes from.
