AWS CEO Says Space Data Centers Are Still Far From Practical Despite SpaceX’s AI Compute Ambitions
Elon Musk and SpaceX may be pushing toward a future where artificial intelligence workloads are powered from orbit, but Amazon Web Services CEO Matt Garman believes space-based data centers are still a long way from becoming practical.
Speaking during a podcast interview, Garman said the idea of building data centers in space sounds exciting, especially as demand for AI computing continues to explode. However, he argued that the economics, engineering challenges, launch requirements, cooling issues, and maintenance problems make large-scale orbital compute infrastructure unlikely to become viable within the next decade or two.
According to Garman, the first major obstacle is launch capacity. Building a serious data center network in orbit would require sending enormous amounts of hardware into space. That includes servers, GPU racks, power systems, thermal management equipment, communications hardware, structural components, and replacement parts.
While SpaceX is working to increase launch frequency with Starship, Garman remains skeptical that launch cadence will improve quickly enough to support massive space data center construction. He suggested that even one launch per day may only be a starting point for making the concept economically realistic.
SpaceX’s vision depends heavily on Starship, its next-generation reusable rocket. The company aims for Starship to eventually deliver around 100 metric tons to orbit per launch when operating in reusable mode. Musk has also discussed ambitious long-term goals involving huge annual payload capacity and future orbital infrastructure designed to support AI computing at unprecedented scale.
One of SpaceX’s proposed ideas is a data center constellation sometimes referred to as Starmind, which could eventually support enormous compute capacity using solar power in orbit. Long term, the company has suggested the possibility of deploying up to 100 gigawatts of compute capacity, though such a project would require major advances in launch economics, space manufacturing, power generation, cooling, and autonomous operations.
Garman pointed out that the weight of modern AI hardware is a serious limitation. A single rack of GPUs can weigh around a ton, and AI data centers require many such racks to deliver meaningful performance. Launching that much mass into orbit would be complicated and expensive, even with a fully reusable heavy-lift rocket.
Power is another challenge. On Earth, hyperscale AI data centers can consume hundreds of megawatts or even gigawatts of electricity. In space, solar panels can generate power, but scaling orbital solar arrays to support gigawatt-class AI workloads would require huge structures. Garman noted that most solar power systems are discussed in kilowatts, while modern AI infrastructure is increasingly moving toward megawatt and gigawatt levels.
Cooling may be an even bigger issue than many people expect. Space is often described as cold, but cooling electronics in orbit is not simple. On Earth, data centers rely on air, liquid cooling, and large-scale thermal systems to move heat away from servers. In space, there is no air to carry heat away through convection. Heat must be managed mainly through radiation, which requires large radiators and careful thermal design.
Garman emphasized that AI servers run extremely hot, and dissipating that heat in space would be difficult. Musk has previously suggested that aluminum radiators could achieve significant thermal radiation performance, but scaling that approach for dense AI compute satellites remains a major engineering challenge.
Radiation is another concern. Servers in orbit would face a harsher environment than hardware in terrestrial data centers. Cosmic rays and solar radiation can damage electronics or cause errors, meaning space-based compute systems would need additional protection, redundancy, and specialized designs. That would likely make them more expensive than standard data center equipment.
Serviceability also presents a major problem. On Earth, cloud providers can repair, replace, and upgrade servers regularly. Technicians can swap failed drives, install new GPUs, improve cooling systems, and maintain power infrastructure. In orbit, every repair is more difficult, expensive, and potentially impossible without advanced robotics or frequent servicing missions.
Despite his skepticism, Garman did not dismiss the concept entirely. He said the technical challenges are likely solvable over time, but he believes the timeline is much longer than some space computing advocates suggest. In his view, orbital data centers may eventually become possible, but not soon enough to reshape the AI infrastructure market in the near future.
The debate comes as artificial intelligence is driving unprecedented demand for compute power. Companies are racing to build larger data centers, secure more GPUs, access cheaper energy, and improve cooling efficiency. This pressure has sparked interest in unconventional ideas, including underwater data centers, nuclear-powered facilities, and now space-based AI compute platforms.
For now, however, terrestrial data centers remain far more practical. They are easier to build, easier to maintain, easier to cool, and cheaper to connect to existing power grids and fiber networks. Space-based data centers may offer advantages in the distant future, such as abundant solar energy and reduced land constraints, but the current barriers remain substantial.
Garman’s comments highlight the gap between ambitious space infrastructure visions and near-term commercial reality. SpaceX may continue working toward lower launch costs and reusable heavy-lift transport, but turning orbit into a competitive home for AI data centers will require breakthroughs across multiple industries.
In short, the future of AI computing may eventually extend beyond Earth, but according to the AWS chief, that future is still many years away.
