A Push Toward The Cosmos

Every time you ask AI to write a line of code, summarize a pdf or write an essay at the last minute, a server somewhere on Earth gets very very hot. 

Electrical utilities are already facing a tidal wave of new demand from the electrification of industry and transport. A solution is needed to keep peace with AI development as overall electricity demand is projected to triple in the coming years. 

Training next-generation models like GPT-6 will require computing clusters that draw up to five gigawatts of power. A facility capable of accommodating that size would exceed the capacity of the largest power plant in the United States. 

Tech companies are exploring a radical, physical solution: shifting gigawatt-scale data centers from the Earth’s surface to space. 

Currently, data centers account for roughly 1% or 2% of the world’s electricity, but according to a Goldman Sachs report, that number could double by 2030 alone.

Moving these facilities into orbit provides a massive power generation advantage. Terrestrial solar farms in the US achieve a median capacity factor of just 24%. To put this in perspective, capacity factor is a metric that compares the actual energy produced by a power plant over a given period against its absolute maximum potential output. Because terrestrial solar panels typically receive 4-6 hours of direct sunlight a day and are subject to weather and changing seasons, they only capture a fraction of what they theoretically could. Conversely, if a data center is parked in a sun-synchrous orbit, it remains in the range of continuous solar illumination, achieving a capacity factor of greater than 95%.  

Earth-bound facilities consume vast amounts of fresh water to prevent servers from melting down. In space, engineers can use large deployable radiators to passively dissipate gigawatts of waste heat directly into deep space, which sit at an average temperature of about -270 Celsius. The European Commission’s ASCEND feasibility study verified that this orbital approach would not require water for cooling, providing a critical advantage in times of increasing drought. 

Terrestrial data also relies on subsea fiber-optic cables that are susceptible to natural disasters such as undersea volcanic eruptions. In contrast, spatial laser links are incredibly difficult to hack or demote. 

In terms of rocket emissions, companies are hoping to decrease the carbon impact of their space-based data centers.  The ASCEND study concluded that to actually reduce the digital carbon footprint, the industry must develop a high-capacity launcher that is overall ten times less emissive than current rockets. Current companies are aiming for economical deployment by 2030.

However, these projects are not without debilitating physical and environmental barriers that actively threaten their viability. The vacuum of space is aggressively hostile to standard electronics; solar flares and cosmic radiation can instantly damage mainstream microchips and sensitive memory components that are not built to work in orbit. Consequently, fixing these inevitable hardware failures in orbit is nearly impossible since fully remote robotic repair technologies do not yet exist, driving up maintenance costs exponentially.

Despite these physical and environmental barriers, the transition to orbital infrastructure is actively underway. Companies like Starcloud have already launched Starcloud-1, the first satellite to run a version of Gemini in space and the first spacecraft to train an LLM — the nano GPT model.

For massive scale infrastructure, ASCEND project aims to deploy one gigawatt of space data center capacity before 2050. These modular space infrastructures would be assembled directly in orbit using robotic technologies from the EROSS IOD program, which is scheduled to fly its first mission in 2026. 

Even though fully remote robotic repair technologies do not yet exist, the aerospace industry is betting heavily on the fact that humanity's data footprint is rapidly outgrowing the planet. As the global grid struggles to sustain artificial intelligence, the next logical step for our digital infrastructure is looking up.