Joseph Flaig
AI, it seems, is too big to fail. A small group of heavily AI-focused companies prop up the US economy, despite significant uncertainty about the technology’s ability to deliver profit. If the bubble bursts, the shockwaves will tear around the world.
The only option, apparently, is growth – and that requires huge amounts of energy, with data centre electricity consumption predicted to double by 2030. Tech leaders (and the UK government) hope that small modular reactors (SMRs) could provide a substantial part of the solution. Most plans are in the early stages, however, and are unlikely to supply meaningful amounts of electricity before the mid-2030s. In the meantime, fossil fuels are making up the difference, adding to pollution and greenhouse gas emissions at a time when the opposite is desperately needed.
Data centres are also spreading over vast tracts of real estate. The computing power needed for complex tasks such as coding and video generation requires large amounts of land – an upcoming Oracle data centre in Doña Ana County, New Mexico, will cover 1,400 acres (5.7km²), for example.
What if there was another option? That is the hope of start-ups including Starcloud, directors at Google, Jeff Bezos and Elon Musk, who are all increasingly looking to space.
“It’s really all about the Sun,” said Musk to World Economic Forum co-chairman Larry Fink on stage in Davos last week (22 January). “That’s why one of the things we’ll be doing with SpaceX within a few years is launching solar-powered AI satellites, because space is really the source of immense power, and then you don’t need to take up any room on Earth – there’s so much room in space – and you can scale to, I think, ultimately hundreds of terawatts [of new solar capacity] a year.”
Such orbital installations would involve giant arrays of solar panels powering specialised AI chips. SpaceX, which operates roughly two-thirds (66%) of all active satellites with its Starlink constellation, knows more than most about the practicalities of taking on such a challenge – but it will still be far from easy.
“The devil is in the detail,” said Professor Matthew Santer, co-director of space, security and telecoms at Imperial College London, to Professional Engineering. While he said that space data centres are “not necessarily a bad idea per se”, he stressed the need for a clear role and application for such concepts.
“Depending on the operating altitude, the communications latency will be more or less severe (although always substantially greater than the terrestrial equivalent), and this has implications for what functions an on-orbit data centre could perform,” he said.
“There is a lot of solar energy potentially available, and a lot of very cold space to radiate excess heat away into, but the on-board thermal management would be potentially very challenging. How would the heat be transferred from the processors to the radiators at a sufficient rate, for example?”
Those issues are a key consideration for Project Suncatcher, a Google Research ‘moonshot’ exploring equipping satellite constellations with its TPU circuits to “scale machine learning compute in space”. Announced in November last year, the project envisions ‘compact constellations’ of solar-powered units connected by free-space optical links, which use light to transmit data.
Likely operating in Sun-synchronous low Earth orbit to access near-constant power, the system would use tight clusters of modular satellites positioned within hundreds of metres of each other. Challenges being investigated include increasing data bandwidth, control of the close formation, radiation tolerance of the TPUs and economic feasibility.
Speaking about the field in general, Santer said that much of the required technology already exists and a data centre would be designed following a similar process to any other large space structure. Any companies envisaging ‘mega-structures’ would need to consider in-space assembly and servicing, however.
“There is a lot of technical development required before this is feasible,” he said. “There is a big issue with service life, solar panel efficiency degrades under UV (ultraviolet light), radiation can permanently affect processors, systems have to be sufficiently robust against debris impact etcetera.”
Launch, in-orbit assembly, radiation shielding and the overall cost are all “major concerns”, he added. “In my own area, the thermal management issues and controlling the structural dynamics – particularly of mega-structures – are big questions. The communications latency and how the data centre will operate with this constraint is also an issue.”
Another company involved in the field is Oxfordshire firm Space Solar, which said it is ready to deliver large, lightweight arrays for orbital data centres, alongside its main focus on developing space-based solar power (SBSP), which would instead harvest sunlight, convert it to microwaves and beam it to receivers on Earth.  
“Despite the growing focus on orbital data centres, not everything that is needed will be optimal in space,” said co-CEO Sam Adlen in a blog post on the company’s website. “Practical latency constraints mean not all compute infrastructure can move to orbit. Orbital data centres make most sense for AI training, space command and control, and Earth observation data fusion, but far less sense for AI inference.”
This will instead be provided by terrestrial data centres, he suggested, which could be powered by SBSP as launch economics evolve and make it a cheaper power source.
Other concerns about space-based AI satellites include the increasing intrusion of manmade objects in astronomical readings, and the growing risk presented by space junk – but as with other AI deployments, it seems likely that environmental concerns will come second to potential profit.
Want the best engineering stories delivered straight to your inbox? The Professional Engineering newsletter gives you vital updates on the most cutting-edge engineering and exciting new job opportunities. To sign up, click here.
Content published by Professional Engineering does not necessarily represent the views of the Institution of Mechanical Engineers.
Please enable JavaScript to view the comments powered by Disqus.
Read now
A weekly round-up of the latest Professional Engineering news, straight into your inbox.
Log in and opt in to receive PE Weekly
Javascript Disabled
Please enable Javascript on your browser to view our news.
Services for home and work
Recognising engineering excellence
Financial help and personal support
Improving the world through engineering
Flexible spaces at our head office in Westminster, London
© 2026 Institution of Mechanical Engineers. IMechE is a registered charity in England and Wales (206882), as well as Scotland (SC051227).

source