AI in Space: Why Tomorrow’s Networks Will Need an Interconnected Orbit

­AI’s growth is pushing the boundaries of what our digital infrastructure can support. In many places, the power required to train and run models is already outpacing the speed at which new energy capacity can be deployed, and the physical limits of terrestrial grids are beginning to shape the debate about what comes next.

At the Saudi Investment Forum last month, Elon Musk outlined a future in which large-scale AI compute is placed in orbit, powered by continuous solar exposure and cooled in the vacuum of space. In his view, the economics could shift within five years, creating future scenarios where solar powered AI satellites – or even compute clusters in deep space – will outperform Earth-based facilities for certain types of workloads. And Musk isn’t alone. Jeff Bezos has spoken about the lunar environment as a potential foundation for energy intensive operations, and Google CEO, Sundar Pichai, recently announced that Google will test machine learning hardware in orbit through its Project Suncatcher initiative.

There’s certainly enough ambition going around, but underneath it is a genuine shift in how the industry is thinking about energy, compute potential, and the long-term evolution of our digital infrastructure stack. The prospect of orbital computation also introduces new dynamics – and dependencies – for the movement of data, the structure of networks, and the location of AI workloads. Space based systems are set to become an essential part of our digital backbone alongside terrestrial data centers, edge locations and high-capacity network hubs. This reinforces something we’ve known for some time – that the future of AI and communications won’t just depend on the technology that facilitates it, but on the infrastructure that supports it. Interconnection, which refers to the high-performance exchange of data between networks, platforms and compute locations, is already becoming an AI necessity. If that AI computation begins to extend outwards and into space, it stands to reason that the interconnection environment holding it all together will need to extend outwards too. There’s little doubt that AI will extend beyond the confines of Earth – what we should be asking is how future systems on the ground and in orbit will be linked to support real-time applications, distributed model training, and the general flow of data on an extra-terrestrial scale.

Power, Cooling, and the Limitations of Earth

Humans are born problem-solvers. The discussion about placing compute in orbit is accelerating because the limitations of terrestrial infrastructure are becoming harder to ignore. AI models require enormous quantities of power, and the growth curve is steepening as organizations scale both training and inference workloads. Many regions are already experiencing constraints on grid capacity, and new data center projects face long development timelines, high energy costs and drawn-out permitting processes. This is the problem, and it’s prompting industry leaders to explore environments where energy supply is more abundant. When you consider the Earth only receives roughly one two-billionth of the sun’s total energy output, there’s a lot left on the table. Continuous solar exposure in orbit will provide a stable source of power, and recent advances in lightweight photovoltaic materials make the concept increasingly credible for large scale deployments.

Cooling is another driver placing space on the industry’s radar. Modern supercomputers are dominated by cooling systems that add weight, cost and design complexity. In orbit, heat can be dissipated through radiation without traditional cooling infrastructure, which opens the possibility of higher density computing without the constraints of water availability or thermal limits – both increasingly problematic on Earth. Companies like Google have already started modelling the economics of space-based systems, and estimate the combination of uninterrupted solar energy and radiation cooling could make certain workloads viable beyond Earth within a decade. While terrestrial data centers will remain essential for proximity, latency (and probably regulatory) reasons, the growing interest in orbital infrastructure shows that we are thinking about tomorrow’s problem today: How will we sustain AI’s energy footprint in a decade’s time, or two decades’ time?

Building the Orbital Layer

If compute does begin to extend into orbit, which is increasingly likely, the next challenge is how these systems participate in the wider digital ecosystem. Satellites cannot operate as islands. They need structured pathways to exchange data with Earth, with each other, and with edge locations that support real time applications. This will require interconnection points, predictable routing, shared technical standards – effectively the same principles that underpin the terrestrial Internet. In practice, an orbital layer would need to function much like today’s peering ecosystems, with Internet Exchanges (IXs) enabling a focus on secure traffic exchange, millisecond latency links, and reliable pathways between numerous independent networks and platforms. That’s where optical, laser-based communications enter the picture. Radio-based satellites are already a useful addition to our connectivity ecosystem, but while good for coverage, they tend to struggle with bandwidth per second throughput, resulting in increased latency due to bottlenecks. And latency is the enemy of AI, particularly when it comes to the kind of close-to-real-time responses needed AI-supported services that interact with the real world. Optical laser-based signals, on the other hand, are able to support the speed, bandwidth and precision needed for AI successful AI implementation. We’ve already pretty much mastered free space optical (FSO) communication on Earth – the next hurdle will be getting it to work in space.

Connecting Earth to Orbit

The inconvenient truth is that FSO still has practical limits that prevent it from supporting consistent, high quality data exchange in environments like space. Laser links can carry far more data than radio signals, but they’re also easily disrupted by the environment they travel through, particularly when that environment includes cloud cover, atmospheric turbulence, and changes in air density that can distort beams and reduce throughput. For satellites that need to hand off data reliably to Earth and vice versa, this remains a major sticking point.

The European Space Agency’s OFELIAS project is currently studying how these problems can be managed through smarter protocols, algorithms, and better handover mechanisms between satellites and ground stations. DE-CIX is currently working with the German Aerospace Centre (DLR) as part of the OFELIAS project to figure out how these optical links could be optimized to successfully communicate with networks on Earth. This goes hand-in-hand with DE-CIX’s broader work on the concept of Space-IX, which looks at what large scale interconnection in orbit might require in the long term, allowing traffic to move between satellites, ground infrastructure and edge locations once space based systems begin to carry meaningful workloads.

Space as Connectivity Backbone

We’ve mastered the ability to put satellites into orbit, and our mastery of AI is levelling up with each passing month. That means the boundary between Earth and orbit will matter far less in the future than the ability to move data intelligently between them. That’s where the real gain lies. If space is to become part of the Internet’s future – and it very much will – the groundwork for that future needs to be built now.