An Off-Planet Solution to an Earthly Problem
The rapid expansion of artificial intelligence is straining global energy grids and resources. Training a single large AI model can consume an immense amount of electricity and water for cooling, leading to an impending energy crunch that threatens to slow
down AI development. This has prompted tech leaders and innovators to look for radical solutions, with some turning their gaze upwards. The concept is no longer science fiction: using space-based assets to support the massive computational needs of AI on the ground. Companies are now exploring placing data centres in orbit, where they can be powered by abundant solar energy and cooled by the vacuum of space, thus bypassing Earth's resource constraints.
The New Players in the AI Race
A new category of companies is emerging at the intersection of space and AI. Startups like Starcloud are making headlines by launching hardware, including powerful NVIDIA GPUs, into orbit with the goal of training AI models entirely in space. In late 2025, Starcloud successfully trained its first AI model in orbit, a significant milestone for the industry. These ventures aim to build constellations of thousands of satellites that function as orbital data centres, serving global cloud providers. They are joined by established aerospace giants like Blue Origin and SpaceX, which are filing plans for massive satellite networks designed specifically for orbital AI computing. Even tech behemoths like Google are getting involved with initiatives like Project Suncatcher, which will test machine learning hardware in space.
It’s Not All About Orbit
While orbital data centres are a key part of the vision, the headline's focus on "ground-based" processing points to a more immediate and tangible application: using satellite networks to improve AI on Earth. This is where companies like Viasat and others come in. Modern AI relies on vast amounts of data, often generated in remote locations far from traditional fibre-optic networks—think industrial sensors, agricultural monitors, or disaster zones. Low-Earth Orbit (LEO) satellite constellations can provide high-speed, low-latency connectivity to these hard-to-reach places. This allows data to be sent quickly to and from ground-based AI processing centres, effectively creating a more resilient and geographically diverse infrastructure that is less dependent on terrestrial cables.
A Hybrid Infrastructure Emerges
The future of AI infrastructure is shaping up to be a hybrid model, combining terrestrial data centres with a robust network of satellites. For ground-based operations, satellites offer a crucial layer of redundancy. If a fibre connection is severed by a natural disaster or other mishap, a satellite link can ensure critical AI workloads remain online. This capability allows companies to deploy edge computing resources in remote areas, processing data locally and using the satellite network to communicate key insights back to a central cloud. For example, a satellite can use on-board AI to analyze images, identify objects of interest like ships or fires, and transmit only the essential data, saving massive amounts of bandwidth. Companies are already using AI on satellites to enable autonomous operations, from navigation to self-correction and inter-satellite communication.
Challenges and the Road Ahead
Despite the immense potential, the path to space-powered AI is filled with challenges. The technology is largely in a demonstration phase, focused on proving that complex computing hardware can survive launch and operate reliably in the harsh environment of space. Launching and maintaining tens of thousands of satellites is an expensive and logistically complex endeavour. Furthermore, these new space ventures are entering a fiercely competitive AI market dominated by established players like NVIDIA, Google, and Amazon Web Services. However, the underlying demand for AI compute is so strong that a new ecosystem is forming, with some companies focusing on launching satellites, others on building the hardware, and a new class emerging to offer in-orbit data services. This convergence of industries points to a future where space is not just a place for exploration, but a critical component of our global digital infrastructure.
