Space data centres: The final frontier of digital infrastructure

Space data centres: The final frontier of digital infrastructure

AI-generated image of a data centre server room in space

Capacity boldly goes where no data centre has gone before: into space

Space: the final frontier. The stars above us may dazzle, but they also offer something increasingly crucial in the form of space. As the demand for computing power on Earth skyrockets, could the vastness of space hold the key to housing the data centres and infrastructure of the future?

Data centres in space may seem like a far-fetched fantasy – an idea for hardware enthusiasts with their heads in the stars. Yet it's a fantasy that could one day become reality, though not in the literal sense of launching a 100,000-square-foot building into orbit like the house in the 2005 film Zathura.

It’s a concept with significant technological potential: powerful hardware orbiting Earth, delivering low-latency computing wherever it is needed. Imagine Starlink, but with servers; Intelsat, but focused on inference. While the idea sounds exciting in theory, the cold, harsh realities of space pose the question of just how feasible the concept is of data centres in orbit.

Subscribe today for free

What is a space data centre?


The short answer is that nobody knows, but placing compute into orbit is, at the moment, just a theoretical concept.

If you ask AI tools like Adobe Firefly to create images of a hypothetical space data centre, you get massive structures floating above Earth filled with hardware akin to a Borg cube from Star Trek.

The reality, however, would be much more mundane, similar to the satellites we see today.

Space data centres would probably involve a series of powerful GPU-like hardware placed on a small module that’s affixed onto a satellite. Add on a few solar panels and a few other hardware bits and bobs like a DPU and an SD-WAN box, and hey presto: you’ve got a small computing unit that can be placed into orbit.

Effectively, a theoretical space data centre would enable the processing of AI and deep learning algorithms to be conducted on board a satellite to reduce data transfer costs and latency.

An advantage of having a data centre in space is that there’s virtually unlimited space in orbit, meaning there’s no limit to the amount of compute you could fire up there. There’s also boundless energy from the sun, and the environmental impact on the ground and among local communities would be greatly reduced.

There’s even the idea of creating a constellation of space data centres communicating to form a larger infrastructure footprint. Add to that the concept of photonic semiconductors, which process light rather than energy, allowing for data to travel at greater speeds over long distances without losing signal strength.

Frontier feasibility

IBM, aerospace company KP Labs and the European Space Agency (ESA) previously looked at the idea of processing data in space. The project focused on mitigating challenges around transporting data from satellites to Earth, with the concept of potentially processing information in orbit.

Nicolas Longépé, an Earth observation data scientist from the ESA, told Capacity, however, that the concept of the space-based data centre was a long-term one, with a timeline of 15 to 20 years.

“The business case is not so evident and there’s a lot of technology that needs to be improved so that it could be realistic,” Longépé says.

He explains that the hardware used in current space missions is around 10 years out of date compared to what’s used for computing on the ground. That means it may take a decade to see a satellite fitted with a GPU like Nvidia’s H100 – and even then, it has to be able to withstand space.

The idea of space data centres stems from the need to process data closer to its source. Longépé describes a potential future scenario: “You could imagine that there will be some spacecraft imaging and then passing the image somehow to processing nodes that will collect everything, and then will do the computation and will do some kind of insight extraction so that, at the end, you don’t have any bottleneck to the ground.”

Longépé elaborates on the concept: “Each satellite acquires a huge amount of data information from sensors with higher resolution. The data collected on board is enormous, and you have some constraints due to the ground station that receives the data coming from the satellite. At some point in the future, we may have some kind of bottleneck.”

The ESA has already made some progress in this direction. In 2020, it launched the ф-sat-1 mission, a CubeSat mission in which a deep learning AI algorithm ran on a GPU directly on board a satellite.

While the CubeSat mission demonstrated the possibility of a space-based data centre in its simplest form, the overall concept is a far-off one in Longépé’s view.

A European Space Agency graphic showcasing the recent Φ-sat-1mission
ESA

Here come the satellites


The likely form factor for the space data centre of the future will therefore incorporate the satellite, which comprises a technology sector that has transformed thanks to innovative new players shaking up the industry like SpaceX’s Starlink.

Toby Robinson, chief of strategy and business development at Avanti Communications, elaborates on Longépé’s thoughts, suggesting the volume of data from satellite Earth observation is “huge and growing exponentially”.

While radiofrequency links are currently being used to break bottlenecks and a move to optical links is also being mooted, Robinson suggests having data centres in space “makes lots of sense conceptually”.

“The data collected would be sent to the data centres orbiting Earth, where the processing could be done, and then only the results would need to be sent back down to Earth, which would be a much smaller data requirement.”

Robinson agrees with Longépé, though, that the concept of the space data centre is probably at the very least a decade away.

An idea Longépé raises that could come to fruition in the near term, however, is that of interconnected satellites: “I think within the next two, three or five years, there will be some concept whereby the satellites will be connected, and one satellite will pass some information to another one with what we call an inter-satellite link.”

The realities of space: Hardware’s worst nightmare


There’s a myriad of reasons preventing computing hardware from being placed into orbit to process data in space, with challenges involving power distribution, thermal management, mechanical robustness and weight considerations.

Hardware suitable for use in space has to overcome electromagnetic radiation emitted by the sun, while also being rugged enough to survive being blasted into space on board a massive rocket and handle incredibly low temperatures.

Ken O’Neill, space systems architect at AMD Xilinx, is one of the minds behind Versal XQR SoCs, comprising embedded computing hardware suitable for use in space.

He acknowledges the potential for data centres in space, but outlined to Capacity the significant engineering hurdles that need to be overcome to make the dream a reality.

“There are so many competing factors that have to be considered when you're making a component selection decision for space,” O’Neill explains. “Size is one, radiation effects is another.”

The Sun
The Sun is harmful to both space-faring humans and hardware

O’Neill highlights the extreme conditions that space-bound hardware must endure: “If you're launching something into space, you’re subjecting the part to a huge amount of vibration. It’s so loud; the vibration is just absolutely shocking.”

The thermal environment in space presents another significant challenge. O’Neill describes how a satellite in low Earth orbit experiences dramatic temperature swings: “It goes through these temperature cycles where it’s being heated by the sun, and then it’s being totally shaded from the sun and just losing all its heat, radiating it out into space. That’s a very thermally challenging environment.”

Power management is another critical consideration. “A big modern satellite might have 20 to 30 kilowatts of power generation coming from the solar panels, and that is challenging the power distribution system inside the satellite,” O’Neill says.

In addition, teams designing hardware for space face tension between device size and mass restrictions enacted by space agencies. NASA can’t exactly fire up an AMD Instinct MI325X Platform of eight GPUs, or even a Cerebras WSE-3, a single chip the size of a dinner plate.

“People want to do more in their programmable logic devices, and one way of doing more is to have bigger and bigger devices – but then, that runs counter to the weight and the physical footprint,” O’Neill explains. “There's no one-size-fits-all kind of solution for a programmable logic device in space.”

Considerations around hardware weight are why AMD offers two sizes in its Versal line: the larger Core 1902, which measures 45 by 45 millimetres, and the smaller Edge 2302, which is 23 by 23 millimetres, about one-fourth the area of the larger unit.

AMD Versal XQR Adaptive SoCs
AMD Xilinx's Versal XQR Adaptive SoCs are purpose-built for performance in space | Credit: AMD Xilinx

“We’re trying to give customers choices so that they can choose an appropriately sized part for the design problem they’re trying to solve,” O’Neill says. “It doesn't mean that every design for a programmable logic device needs to be the biggest, baddest, highest performance thing that we can offer.

“There’s definitely still a market there for devices that have a small form factor and don’t need to do a whole heck of a lot, but still have work to do – and they just need to sit in the corner and do their thing.”

Yet another hurdle is thermal management in the vacuum of space. O’Neill explains: “What do you do with 20 or 30kW of heat when you're surrounded by a vacuum? There's no air for cooling fans to blow over heat sinks.

“The only thing you can do is to conduct heat away from integrated circuits into the mechanical chassis of the satellite itself, and then radiate the heat out into space.”

Despite the engineering minefield that is getting hardware to function in space, O’Neill remains optimistic about the future of space-based computing.

“Semiconductor companies like ourselves are looking for ways to do more processing in more power efficient ways with every architecture and every process generation, and it is driving the research that we’re doing here going forward.”

Edge of the Edge


The entry into orbit of Sputnik 1, the first artificial Earth satellite, in 1957 signified a landmark moment in the space race. Similarly, one could argue that HPE’s Spaceborne Computer was the ‘Sputnik 1’ of space data centres, pioneering the concept and setting the stage for future advances.

The Spaceborne Computer was essentially a commercial off-the-shelf piece of hardware fired on board a rocket simply to see if it would survive the journey into space.

The project was so successful that a year and a half into its lifespan, it was still operating – and it got NASA excited.

Now, HPE is working to improve things with Spaceborne Computer-2, in what could be considered a proto-space data centre.

HPE's Spaceborne 2 aboard the ISS
HPE's Spaceborne 2 aboard the ISS | Credit: HPE/NASA

Mark Fernandez, the principal investigator for Spaceborne Computer-2, told Capacity that the unit houses similar hardware to what’s found in data centres on Earth, with the added bonus that it can function at what is truly the edge of the edge.

Having returned to the International Space Station in January, the unit has been processing high-performance computing and AI workloads in space.

Equipped with KIOXIA 960GB RM Series SAS drives, 1,024GB NVMe drives and 30.72TB PM6 Enterprise SAS SSDs, Spaceborne Computer-2 houses some serious hardware for use in space.

The computing unit is capable of handling massive amounts of data processing and storage in an extremely harsh environment Spaceborne Computer-2 even survived an aurora borealis event that was one of the biggest radiation events in 500 years, coming out with just four software issues that were easily corrected.

The second iteration of Spaceborne was overhauled to offer significant improvements, including switching from AC to DC power to make use of solar energy, integrating water cooling to reduce heat by around 80% and incorporating GPUs to power intense workloads – including AI and machine learning.

Fernandez suggests that Spaceborne shows how a mentality shift is needed to demonstrate that this concept can work: “I can talk to people about how fast the networks will be and how much better it will be when we switch from radio to light, et cetera. But you have to change your mentality to say, ‘Why am I doing this? What nugget of insight or piece of information do I want out of this massive amount of data?’ And let me get that if it’s there, and that will be far smaller than the raw data.”

The proto-space data centre that is Spaceborne Computer-2 could prove a vital tool for agencies like NASA as humans return to the moon in the next decade.

Hardware offerings like HPE’s space system can be used to perform high-resolution multi-spectral analysis to find water on or just beneath the moon’s surface.

The moon is just the beginning, with systems like Spaceborne Computer-2 potentially playing an important role in humanity’s greatest space mission: Mars.

Mars

Fernandez suggests such systems could play a role as a “gateway” – an orbiting satellite that could process and optimise parts of data, rather than having to send information all the way back to Earth, which would take time and massive amounts of power to perform.

“You would go from the gateway to the moon and back, and from Earth to a gateway and back,” Fernandez explains. “You can optimise from Earth to the gateway, and then optimise from the gateway to the moon.

“Suppose on the gateway we put a data centre: we could use 5G from the moon up to the gateway, up to your orbiting data centre to help you out. We envision a similar possibility when we get to Mars.”

Back to reality


Space data centres are a concept that, for now, appears a distant yet fast-approaching reality.

Proto-projects like the ESA’s ф-sat-1 and HPE’s Spaceborne Computer-2 show there’s potential for data centres to be housed in space, even if we have to wait a few years to see them.

Fernandez outlines a vision of the future in which space data centres operate in three distinct locations: aboard lunar rovers, on surface-based stations like ‘power towers’ on the moon and in orbit on space stations.

“Our vision is to provide orbital, surface and mobile data centres as a service, allowing scientists to focus on their instruments and sensors while we handle the heavy computational lifting, even outpacing earthbound bandwidth capabilities,” Fernandez says.

While challenges remain, the foundations are already being laid to bring every hardware nerd’s dream to reality – and, with it, a future in which space becomes not just the final frontier but also a hub of advanced digital infrastructure.

This article first appeared in Capacity's December 2024 Magazine

RELATED STORIES

Astronauts to test revolutionary quantum communication system in space

Sky's the limit: How hybrid satellite solutions are reshaping connectivity

Gift this article