Are orbital data centres the answer to AI compute demand?

As energy becomes the most sought-after resource of the technological age, does it make sense to relocate data centres into low-Earth orbit (LEO) where solar energy is almost limitless?

The richest man in the world certainly thinks so. For a science-fiction devotee like Elon Musk, the constraints of terrestrial infrastructure in delivering the compute power required to advance AI have naturally turned his attention to space.

On 30 Jan, Musk’s SpaceX made an application with the US Federal Communications Commission (FCC) to launch an “orbital data centre system” using a constellation of up to one million solar-powered satellites. The filing describes moving compute infrastructure into space as the “only logical solution”.

In February, Musk merged xAI into SpaceX to create a $1.25trn company with ambitions to develop orbital AI infrastructure. And SpaceX’s much-anticipated IPO on 12 June is an important fundraising milestone for Musk’s wider orbital AI strategy and plans for space-based data centres.

By combining his AI company with SpaceX’s space launch capability and Starlink’s satellite networks, Musk has placed himself in a position of owning both the demand and supply side of future orbital computing platforms. And for investors in the wider tech community, Musk’s orbital ambitions have led many to ask whether he can do for data centres what he did for electric vehicles—that is, to accelerate the development of an entirely new market.

What are the benefits of orbital data centres?

The data centre business has entered a golden age of unparalleled investment on the tailwinds of AI advances. GlobalData predicts that the global enterprise data centre and hosting market will grow to $188.2bn by 2029, with a compound annual growth rate of 10.9% between 2024 and 2029.

The investment surge is largely due to the AI boom, as well as a longstanding shift away from on-premises data centres. Colocation services are set to grow rapidly, making up 47% of the total hosting market by 2029, surpassing the application hosting and data centre services segments.

But the backlash from community and environmental groups concerned about impact to energy and water supplies is quickly gaining momentum and protests outside proposed data centre construction sites are becoming common place.

The case for developing orbital data centres focuses largely on overcoming these terrestrial constraints. Uninterrupted and unfiltered sunlight in space could provide a source of abundant renewable energy to meet AI compute needs. For context, solar panels operate at around 30% efficiency when accounting for weather and sunlight hours on Earth.

Data centre air conditioning and liquid cooling systems are one of the highest operating costs for data centres, as well as placing high demand on local grids. The space vacuum of minus 270 degrees Celsius has potential cooling advantages, although heat can only be dissipated through radiation, making thermal system development and management a significant engineering challenge.

There are also potential latency benefits. Constellations of satellites could provide orbital data centres with a more direct intercontinental communications and cloud services than traditional fibre networks, reducing latency. Furthermore, without zoning and grid constraints, and regulatory approval, data centres could be developed very quickly compared to the average 18 months to two years it takes to develop a terrestrial data centre.

All of the benefits have to be weighed against engineering challenges that cannot be underestimated. The sheer technical complexity and cost of launching hardware into orbit is still prohibitively expensive, even accounting for SpaceX’s reusable rockets.

Other major hurdles include maintaining, servicing and upgrading equipment in space where hardware cannot easily be accessed. Added to which, the equipment is not as durable in space as it’s exposed to cosmic rays and harsh conditions.

What is driving orbital data centre investment?

Since 2021, when SpaceX began its ride share space launches, there has been an explosion in the frequency of space launches. Commercial space travel and the advent of SpaceX’s pioneering reusable spacecraft have rapidly reduced the cost of each trip, creating the conditions for data centres to move from theory to practical possibility.

There are two distinct demand drivers for orbital data centre development, according to Daniel Thorpe, head of data centre research at commercial real estate and investment management company, Jones Lang LaSalle. The first is the long-term terrestrial limitation of energy supply and available land as the demand for AI compute power increases exponentially.

The second, a more short-term driver, is an immediate need for compute power in space. The proliferation of space activity, such as observation satellites that collect terabytes of raw data per day means that there is currently a data bottleneck in space. Sending raw data down to Earth for processing is slow and expensive, making data centres in space for orbital processing a much better option.

“Those demand drivers will both increase as we start to see more autonomous space operations coming through,” says Thorpe.

AI native companies, in particular, are watching the orbital data centre space with interest. The current pace of technology development makes planning a challenge for any company trying to anticipate AI compute demand. Thorpe likens orbital data centre discussions to the rhetoric around quantum computing which has, at times, become hyperbolic.

He argues that the approach to data centres in space should not be binary: both hype and laissez-faire should be avoided. However, the fear of being left behind, as well as technology obsolescence risks, will always weigh heavy on technology leaders. Just as the advent of quantum computing threatens to topple today’s classical computing cyber security guardrails, investing in terrestrial compute power could prove to be a fruitless strategy if the future of compute power lies in space.

But Thorpe urges tech leaders not to focus on the terrestrial versus space dichotomy. The idea that every data centre needs to be AI enabled or have the latest GPU chips rather than CPU chips is simply not true. There will always be a need for traditional terrestrial data centres which Thorpe describes as “the bread and butter” of most of our applications that don’t need the latest technology.

Orbital data centres won’t necessarily run standard compute systems more efficiently, or effectively. They come into their own for specialised use cases, says Thorpe. And that would have enormous implications for certain sectors like life sciences and healthcare.

Orbital data centre players

Whether orbital data centres ever become economically viable remains uncertain, but a growing ecosystem of private sector companies, research initiatives, and governments are all testing prototypes and exploring the future of compute power in space.

According to GlobalData forecasts, the space economy market was worth $421bn in 2024 and will reach $511.2bn by 2029, growing at a compound annual growth rate of 4% over the five-year period.

Axiom Space and Starcloud are first mover companies in the commercial orbital data centre market. In a white paper, “Why we should train AI in space”, Starcloud predicts that orbital solar energy could become more cost effective than terrestrial power, with the presumption that launch and deployment costs continue to reduce. The white paper outlines that for a 40 MW cluster operated for ten years in space versus on Earth, they estimate a $138m potential saving on energy.

The future of orbital AI infrastructure

Jason Aspiotis has been a true believer and evangelist for orbital data centres since 2021, when he first started Axiom Space’s orbital data centre business. He says he did this with the conviction that the sector would become a real market by the end of the decade.

“I agree, to an extent, with Elon Musk and others’ vision of our civilisation’s compute and AI infrastructure migrating to space, given unlimited power, cooling and real estate in orbit and beyond. But I also believe in the right economic sequencing of technology versus market demand and adoption,” says Aspiotis who has been advocating and working on a more tempered approach.

By this he means focusing early orbital data centre capabilities on early adopters, namely military, civil, science, and commercial users that require near real-time analytics and intelligence from their space data. Aspiotis’s timeline for this being 2026-2030.

The next phase from 2030-2035 would be to expand capability to support other emerging use cases like in-space cyber and post-quantum security services, convergence of AI and NTNs, and global sovereign orbital cloud services.

From 2035 onwards, as the technology and launch services further commoditise through Starship and other launch vehicles, and the unit economics start making more sense compared with terrestrial cloud and AI infrastructure, the industry should start addressing terrestrial user needs in aviation, maritime, disconnected environments, critical infrastructure, and eventually, via the proliferation of D2D capabilities, orbital cloud and AI directly to user endpoints such as mobile phones.

Given the initial use cases are somewhat niche, Aspiotis foresees opportunity for smaller players to create value-added solutions at reasonable economics and therefore create small but hard to penetrate market wedges.

Furthermore, when it comes to sovereign orbital cloud and AI capabilities, especially in Europe and elsewhere in the world, SpaceX and other US companies will face challenges with widespread adoption of their capabilities. So even though SpaceX and perhaps hyperscalers have the advantage of scale, smaller players in the global race towards orbital compute may have the advantage of sovereign backing and independence from the US.

How do companies address digital sovereignty in space?

With all the uncertainty around timelines and, indeed, whether the orbital data centre industry will deliver on its promise of abundant compute power, what should enterprise technology leaders be considering now—even if compute power in space has yet to reach boardroom discussions?

GlobalData principal analyst, Isabel Al-Dhahir, echoes Aspiotis’s view that sovereignty will become an increasingly important factor in how the industry evolves. Orbital data centres pose a dilemma to their end users who will want clarity on which jurisdictions their data is being stored under.

“With digital sovereignty becoming a theme of growing importance globally, nations are racing to secure control over their digital infrastructure including data centres. The pressure is mounting as geopolitical fragmentation continues to worsen, and the fear of foreign interference becomes more likely,” says Al-Dhahir.

The US remains especially dominant in both the software and the hyperscale data centre market, granting it disproportionate leverage in global affairs, for example through the credible threat of disrupting, or even outright removing, access to essential services, notes Al-Dhahir.

Sovereign cloud infrastructure, a key part of the wider digital sovereignty agenda, aims to reassure businesses and organisations that their data is secured and governed exclusively by the laws and regulations of a defined jurisdiction. But Al-Dhahir notes that orbital data centres could undermine these assurances.

“Considering that no country owns space, determining which legal regime applies to data stored and processed in orbit risks becoming a regulatory quagmire. End users will likely want control over whether their data is stored terrestrially or in the legal grey zone, but it remains unclear how much choice service providers will offer in practice,” she adds.

Digital sovereignty in space is still a moot point. How the space economy develops over the next decade remains uncertain, but technology leaders, particularly AI native companies will be watching closely.