Amazon Web Services (AWS) CEO Matt Garman has questioned the commercial viability of building artificial intelligence data centres in space, arguing that the technology is unlikely to become economically practical within the next 10 to 20 years. Although orbital facilities could offer access to abundant solar energy, Garman said the cost of launching equipment and the engineering challenges involved make the concept difficult to implement at scale.
Garman shared his assessment during an interview on the Sources with Alex Heath podcast, released on November 10, 2026. His comments come as major technology companies and space businesses explore ways to move AI computing infrastructure beyond Earth. The growing demand for computing power, electricity and cooling has encouraged companies to investigate orbital data centres as a possible long-term alternative to conventional facilities.
Why AWS CEO Matt Garman Is Skeptical About Space Data Centres
Garman acknowledged that placing data centres in orbit has some attractive advantages, particularly the ability to access solar energy. However, he argued that the practical difficulties outweigh those benefits for the foreseeable future.
One of the biggest obstacles is the frequency and cost of rocket launches. Modern AI data centres require enormous quantities of servers, processors, networking equipment and supporting infrastructure. Transporting this hardware into orbit would require a launch system capable of moving large payloads regularly and economically.
Garman pointed out that the global space industry does not currently conduct launches at the frequency that would be required to support a large network of orbital computing facilities.
He compared the pace of space launches with AWS’s own infrastructure expansion. According to Garman, the company is installing one GPU in its data centres approximately every 14 seconds. Moving hardware into orbit at a comparable scale would require a substantial improvement in launch capacity and logistics.
He described orbital data centres as highly unlikely to become economically feasible or practical within the next decade or two.
Cooling Is a Major Engineering Challenge in Space
Beyond launch costs, Garman identified heat management as another significant obstacle.
AI processors generate considerable heat while training models and handling inference workloads. On Earth, data centres use combinations of air conditioning, fans, liquid cooling and other systems to maintain safe operating temperatures.
Space presents a different environment because there is no surrounding air to carry heat away through conventional airflow. Instead, orbital systems must transfer heat through specialised equipment, including radiators that release thermal energy into space.
Garman said dissipating heat in space is difficult and that radiation creates additional complications. Electronic components would need protection against the harsh conditions encountered in orbit.
He nevertheless described these obstacles as technically solvable. His argument is that solving them at a cost that makes commercial sense will take considerable time.
The cooling problem is not unique to AWS’s assessment. Google has also identified thermal management as a central research challenge in its Project Suncatcher initiative, which is exploring whether AI computing hardware can operate in orbit.
Space Data Centres Could Be More Expensive to Build and Maintain
The economics of orbital infrastructure extend beyond the initial launch.
A conventional data centre can be connected to terrestrial electricity grids, fibre-optic networks and established maintenance services. Engineers can replace failed equipment, upgrade servers and expand buildings without sending hardware into space.
Orbital facilities would require specialised spacecraft, radiation-resistant components, thermal management systems and reliable communication links. Replacing damaged equipment could also be considerably more complicated than servicing a server on Earth.
These requirements could increase both construction and operating costs.
The scale of AI infrastructure adds another layer of difficulty. Large data centres contain thousands of processors that need to communicate with one another at high speeds. Replicating that interconnected computing environment across multiple moving satellites would require advanced networking technology.
For companies operating large AI models, the economics would depend on whether the benefits of solar power and orbital deployment could offset the costs of launching, operating and maintaining the infrastructure.
Elon Musk and Jeff Bezos Have Expressed Interest in Orbital Data Centres
Garman’s comments contrast with the ambitions of several prominent technology entrepreneurs who see space as a potential destination for future AI infrastructure.
Elon Musk has promoted the idea of using SpaceX’s launch capabilities to build a large constellation of satellites that could operate as orbital data centres. The concept is intended to address the growing electricity requirements of AI computing by moving some of the infrastructure away from terrestrial power constraints.
Amazon founder Jeff Bezos has also expressed interest in space-based computing. However, Bezos has previously cautioned that some of the timelines being discussed for the technology may be overly ambitious.
The interest reflects a wider concern across the technology industry: the rapid expansion of AI is increasing demand for electricity, land, cooling systems and data-centre capacity.
Moving computing infrastructure into orbit could eventually reduce some of these pressures, but the technology still needs to overcome significant engineering and financial barriers.
Google Is Already Testing AI Chips in Space
While Garman is sceptical about near-term commercial deployment, Google is actively researching the technology through Project Suncatcher.
The initiative is designed to investigate whether Google’s Tensor Processing Units (TPUs), which power AI workloads, can operate reliably in low Earth orbit.
Google’s initial test mission involves a prototype satellite carrying TPU hardware. The company is studying how the processors handle launch vibrations, radiation exposure and extreme thermal conditions.
Cooling is a particularly important part of the experiment. Because conventional airflow systems cannot operate in a vacuum, Google is testing heat pipes and radiators to transfer heat away from its processors.
The company is also investigating high-bandwidth laser communication links that could allow multiple satellites to exchange data and operate together as a computing cluster.
Google has described Project Suncatcher as a long-term research effort rather than a commercially operational orbital data-centre network. Its early experiments are intended to identify technical problems and establish whether the concept can be developed further.
The initiative demonstrates that research into space-based AI infrastructure is progressing, even though large-scale deployment remains uncertain.
AWS CEO Sees Deserts as a More Practical Alternative
Garman also discussed other potential locations for future data centres, including underwater facilities and remote desert regions.
He expressed scepticism about underwater data centres, suggesting that they would face their own technical and operational challenges.
Deserts, however, appeared more promising in his assessment. They offer large areas of available land, strong solar energy potential and environmental conditions that could support certain cooling approaches.
Building data centres in remote areas could also create economic trade-offs for nearby communities. Garman argued that locating facilities too far from population centres could prevent towns from benefiting from high-paying jobs, investment and additional tax revenue.
His comments reflect AWS’s preference for expanding terrestrial infrastructure rather than relying on unproven orbital or underwater alternatives in the near term.
The Bigger Picture
The debate over space-based data centres highlights the growing infrastructure requirements of the AI industry. Companies developing increasingly capable models need more processors, electricity and cooling capacity. This is creating pressure to find new locations and technologies that can support future computing demand.
Orbital facilities could eventually offer access to near-continuous sunlight and reduce dependence on some terrestrial resources. However, the technology must overcome launch economics, radiation protection, thermal management, networking and maintenance challenges before it can compete with conventional data centres.
The difference between Google’s research programme and Garman’s assessment is also important. Testing a small number of AI processors in orbit is not the same as operating a commercial facility capable of supporting large-scale workloads. The transition from experimental hardware to economically viable infrastructure could require years of further development.
Looking Ahead
Space-based data centres are likely to remain an area of research and experimentation as technology companies look for ways to support the next phase of AI growth. Google’s Project Suncatcher and the ambitions of SpaceX demonstrate that interest in orbital computing is continuing, but neither establishes that large commercial networks are ready for deployment. Improvements in launch costs, satellite manufacturing, cooling and communications will be essential to making the concept viable.
For now, Garman believes conventional data centres remain the more practical option, with deserts offering a potentially attractive location for future facilities. His 10-to-20-year assessment does not rule out computing in space altogether; it highlights the gap between technical possibility and commercial feasibility. Whether that gap narrows will depend on advances across both the space and semiconductor industries.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.



