Mini Series: Orbital Data Centers The Future of AI Infrastructure
Watch Video: Orbital Data Centers The Future of AI Infrastructure
AI Is Exploding — And Earth’s Infrastructure Is Hitting Its Limits
AI is growing at an unprecedented pace.
But the infrastructure powering it is starting to crack.
Between now and 2030, more than $5.2 trillion is expected to be spent building data centres globally to meet AI-driven demand. For years, the conversation focused on chips — GPUs, accelerators, and silicon supply chains.
That’s no longer the real bottleneck.
Today, the constraints are power, land, water, and grid capacity.
Meeting demand for hundreds of gigawatts of new energy this decade is an enormous challenge, especially as the most suitable locations on Earth are already filling up. Grid upgrades take years. Power approvals take longer. And communities are increasingly pushing back on land and water usage.
This is where a radical new idea enters the picture: orbital data centres.
Why Space Changes the Energy Equation
In orbit, solar power is near-constant.
No weather.
No cloud cover.
No night cycles.
Just continuous, predictable energy generation.
While terrestrial solar suffers from intermittency and grid congestion, space-based solar can deliver power with unmatched reliability. For energy-hungry AI workloads that run 24/7, that stability matters.
But energy is only part of the story.
The Real Unlock: Cooling
On Earth, cooling is one of the biggest constraints on data centres — and one of the least discussed.
A typical large data centre can consume around 1.1 million litres of water per day, primarily to keep servers from overheating. That’s roughly equivalent to the daily water usage of 1,000 homes.
As AI workloads intensify, water usage is exploding right alongside power demand.
In space, this problem largely disappears.
Space is essentially a vacuum — there’s no air to trap heat. Excess thermal energy can be radiated away naturally, without water, chillers, or cooling towers.
That removes one of the largest operating costs and one of the hardest physical limits facing AI compute on Earth.
A New Compute Layer Above the Planet
Orbital data centres wouldn’t look like traditional server farms.
Instead, they would operate as networks of satellites, connected by high-speed optical links — transmitting data using light rather than electricity. These links allow massive data throughput with minimal latency and energy loss.
Together, these satellites could function as distributed, hyperscale compute nodes in orbit.
Importantly, they wouldn’t replace Earth-based data centres.
They would complement them — adding a new layer to the global compute stack where energy and cooling constraints are radically different.
The Big Players Are Already Moving
This isn’t science fiction anymore.
SpaceX, now merging more closely with xAI, is building what looks like a vertically integrated AI platform — from launching satellites, to orbital infrastructure, to running AI models. Launch, compute, and intelligence under one roof.
At the same time, startups like Starcloud are going all-in on orbital compute. Their first demonstrator satellite is already in orbit. A second is expected later this year, with more planned.
Capital, talent, and momentum are starting to converge.
The Risks Are Real — But So Is the Payoff
Orbital data centres come with serious challenges:
Radiation exposure
Thermal engineering at scale
Orbital debris and congestion
Regulatory uncertainty
Launch and deployment costs
None of these are trivial.
But neither was building the global internet, hyperscale cloud infrastructure, or semiconductor supply chains.
Why This Matters
If AI demand continues to accelerate — which it will —
and Earth-based power and water constraints continue to tighten — which they are —
then space-based data centres could become a critical pillar of global AI infrastructure.
Not a replacement for Earth.
But an extension of it.
In the next phase of AI, the limiting factor won’t be intelligence.
It will be physics.
And physics looks very different in orbit.