Should the cloud leave Earth? Inside the push for data centres in space

The AI boom is colliding with an old, earthly problem: there is not enough power, water or space to sustain it.

With AI models growing larger and more demanding, the warehouses that run them are spreading fast, drawing electricity by the gigawatt and water by the millions of gallons. Faced with rising costs, community resistance and climate pressure, parts of the technology industry are looking upward — quite literally.

What if the next generation of data centres does not sit on land at all? What if they orbit the Earth, or even operate near the Moon, powered by constant sunlight and cooled by the vacuum of space?

AI, more than any previous technology, is pushing this question from science fiction into strategic planning.

AI's energy appetite

Data centres already consume a significant share of global electricity. That share is rising quickly as AI workloads scale. Training and running large language models requires dense clusters of processors that operate continuously and generate intense heat. Cooling them has become as big a challenge as powering them.

On Earth, the options are narrowing: large data centres require vast plots of land, robust grid connections and reliable water supplies. In many regions, those resources are stretched. Local opposition is also growing, as communities question why energy-hungry facilities should raise utility prices and strain infrastructure.

Industry forecasts suggest that demand for data centre capacity will surge sharply through the next decade. AI is the main driver here; even aggressive investments in renewable energy may not be enough to keep pace.

That reality has prompted engineers and investors to ask whether the planet itself has become the bottleneck.

Why space suddenly looks practical

Space offers two advantages that Earth cannot.

The first one is energy. In the right orbit, solar panels receive near-constant sunlight, without clouds, seasons or nightfall.

The second is cooling. Heat can be radiated directly into space, without water or air.

Advances in satellite manufacturing and reusable rockets have lowered the barriers. Launching hardware into orbit is still expensive, but costs have fallen dramatically over the past decade.

At the same time, AI workloads are changing. New research suggests that scaling inference, rather than model size, may deliver greater efficiency. Inference tasks are easier to distribute across many smaller compute nodes, including those linked in orbit.

This shift has encouraged proposals for orbital "clouds" made up of interconnected satellites. Some companies envision clusters flying in close formation, linked by high-bandwidth optical connections. Many others see data storage in space as a secure backup, isolated from natural disasters and cyber attacks on Earth.

Early experiments

The idea of space-based data centres is no longer theoretical. Several companies and governments have begun testing space-based computing systems. Small data processors have been launched into low Earth orbit.

China has deployed the first satellites of a planned in-orbit computing constellation. European researchers have completed feasibility studies backed by public funding.

Private firms are also moving ahead — some are targeting specialised uses, such as processing satellite imagery in space to reduce delays. A few other companies are focused on secure data storage beyond terrestrial networks.

Google, for example, is exploring how clusters of solar-powered satellites equipped with its own AI chips could one day function as a distributed orbital cloud. Its research effort, known as Project Suncatcher, is focused on linking satellites through high-speed optical connections so they can work together much like a terrestrial data centre.

Amazon has shown interest in similar concepts by examining in-orbit processing of data from its Earth-observation satellites and reducing reliance on ground infrastructure. Microsoft has also invested in research on space computing and secure data relay networks that could support future off-planet data handling.

At the same time, startups such as Vast, Ortelio and Xenesis are pushing experimental designs for lunar and orbital computing hubs.

Formidable obstacles

Despite the enthusiasm, scepticism around the technology is widespread. Launching heavy computing equipment into orbit remains costly. For space-based data centres to compete with those on Earth, launch prices would need to drop far below current levels.

AI is forcing a rethink of where computation happens and who bears its environmental cost. Space has entered the conversation not because it is glamorous, but because the alternatives are running out. The race to space is less about escaping Earth than about sustaining the digital world built surrounding AI.

Engineering challenges are equally severe, as most commercial chips are not designed to withstand prolonged radiation exposure. Additionally, cooling systems in a vacuum are complex and bulky. Hardware failures are hard to fix remotely. Even optimistic designs assume regular replacement cycles, adding to cost and risk.

There are also environmental trade-offs. While orbital data centres could reduce emissions on Earth, rocket launches produce pollutants at high altitudes, where they linger longer.

Space debris is another concern. Adding hundreds or thousands of satellites increases the risk of collisions that could disrupt critical space infrastructure.

A future 'above'

At this point, space-based data centres can be best seen as a complement, not a replacement to the existing ones. Hybrid models are more likely in the near term — some workloads may move off-planet, while core computing remains grounded. For businesses that rely on AI rather than build infrastructure, the immediate impact will be limited.

AI is forcing a rethink of where computation happens and who bears its environmental cost. Space has entered the conversation not because it is glamorous, but because the alternatives are running out. The race to space is less about escaping Earth than about sustaining the digital world built surrounding AI.