By
Stuart Kerr, Technology Correspondent,
LiveAIWire

Electricity demand from the world’s data centres is on course to
more than double to around 945 terawatt-hours by 2030, roughly what Japan
consumes today, according to the International Energy Agency’s Energy
and AI report. AI is the single biggest driver of that surge. That
collision between computing and the power grid is why some of the industry’s
loudest voices, Elon Musk among them, now argue the only place left to build
is orbit.

Why the Grid Can’t Keep Up With AI

The problem is physical, not theoretical. The IEA’s executive
director, Fatih Birol, called AI “one of the biggest stories in the
energy world today,” and his agency projects that AI-focused data
centres will more than quadruple their electricity use by 2030. Grids built
over decades cannot absorb that kind of demand in a handful of years, and the
IEA warns that around a fifth of planned data centre projects could face
delays from grid bottlenecks alone.

The squeeze is sharpest in the advanced economies that host most
of the world’s computing. The IEA expects data centres to drive more than a
fifth of all electricity demand growth in those countries to 2030, reversing
decades in which their power use barely moved. That is the kind of pressure
that forces planners to consider options once dismissed as absurd, because
every gigawatt that cannot be connected on the ground is a gigawatt the industry
will look to find somewhere else.

That scarcity is the backdrop to every exotic proposal now
circulating, from small nuclear reactors to floating offshore campuses. Space
is the most radical of them, and until recently it was treated as science
fiction rather than strategy. The strain on real communities from this
build-out is something we examined in what
happens when AI meets the grid.

Musk’s Satellites Become the Launch Pad

Musk’s relevance here is concrete rather than rhetorical. His
company SpaceX already operates the largest satellite constellation ever
built and has filed for authorisation to expand Starlink toward a scale
measured in the hundreds of thousands and, in its long-range ambitions,
beyond. That same launch capacity is what makes orbital computing thinkable,
because the cost of putting hardware into space has fallen further and faster
under SpaceX than under any operator before it.

The link is direct. The first serious attempt to run AI hardware
in orbit reached space on a SpaceX rocket, and Musk has publicly argued that
the economics of space-based AI compute could undercut ground-based data
centres within a few years. Whether or not that timeline holds, the
constellation and the launch business have turned a thought experiment into
an engineering programme.

The First Data Centre in Orbit Is Already
Running

This is no longer a slide deck. In November 2025 the startup
Starcloud launched a satellite carrying an Nvidia H100 graphics processor on
a SpaceX rocket, and weeks later reported training and running an AI model in
orbit, according to Nvidia.
Philip Johnston, Starcloud’s cofounder and chief executive, told the company
that in space “you get almost unlimited, low-cost renewable
energy,” the pitch behind a planned orbital data centre of several
gigawatts.

Starcloud is not alone. Google has outlined Project Suncatcher, a
plan to fly solar-powered satellites carrying its own AI chips, with
prototype launches targeted for 2027. The appeal is the same in every case:
constant sunlight, the cold of space for cooling, and no neighbours to object
to the power draw. The race to power these systems on the ground is the
subject of Google’s
nuclear bet to power AI.

The ambitions are enormous relative to what has flown so far.
Starcloud has described a planned orbital data centre running to several
gigawatts of capacity, fed by solar and cooling panels measured in kilometres
across, a structure with no precedent in orbit. Set against a single H100
chip circling the Earth today, that is a vast leap, and the gap between the
demonstration and the dream is where most of the doubt lives. Yet the field
is suddenly crowded with serious money, from chipmakers to search giants, and
competition of that kind tends to compress timelines that once looked
fanciful.

The Catch: Launch Costs, Heat, and Debris

The obstacles are as real as the promise. Independent experts
cited by the IEA and others note that orbital data centres only make economic
sense if launch costs keep falling sharply, and that radiation, the
difficulty of cooling in a vacuum and the impossibility of sending an
engineer to swap a failed part remain unsolved at scale. There is also the
problem the industry rarely volunteers: more hardware in orbit means more
congestion and more debris in an already crowded low Earth
orbit.

None of this makes the idea hopeless, but it does puncture the
cleaner version of the story. A data centre in the sky is not automatically a
green one, and the environmental maths depends heavily on how the rockets
that build it are fuelled and how often they fly.

There is a sovereignty question too. A data centre overhead does
not sit neatly inside any one country’s laws, and who controls the data
passing through it, who can switch it off, and whose rules apply when
something goes wrong are all unsettled. For governments already nervous about
where their citizens’ information lives, moving it into orbit trades one set
of anxieties for another. These are not reasons the idea will fail, but they
are reasons it will arrive slowly, hedged with regulation, rather than in a
single triumphant launch.

What This Means for You

For most people the immediate stake is the electricity bill and
the grid. The IEA expects data centres to account for more than a fifth of
all electricity demand growth in advanced economies to 2030, which is why
your local energy prices and planning rows are increasingly tied to where the
world decides to put its computing. If even a fraction of that load
eventually moves off-planet, the pressure on terrestrial grids, and on the
communities living next to them, eases. If it does not, the fight over land,
water and power near data centres only intensifies. The fuller picture of
that pressure sits in our report on the
energy crisis of AI.

The Sky Is a Long Shot, but No Longer an Impossible
One

Orbital data centres remain a bet, not a certainty, and the honest
position is that the first gigawatt-scale facility in space is years away and
may never pencil out. What has changed is that serious companies are now
spending real money to find out, and a working AI chip has already trained a
model above the atmosphere. The next test launches in 2027 will tell us
whether the data centre in the sky is the future of computing or an expensive
detour. Either way, the reason the question is being asked at all is that the
grid on the ground is running out of room, and that pressure is not going
away. The likeliest near-term future is a hybrid one, with the heaviest and
most heat-tolerant workloads tested in orbit while everyday computing stays
on Earth, and the balance between the two will shift only as launch costs and
hard engineering allow. The sky is now part of the conversation about where
AI lives, and a few years ago it was not.

About the Author

Stuart Kerr is Technology Correspondent at LiveAIWire, covering
artificial intelligence, cybersecurity, and the social impact of emerging
technology. He publishes daily at LiveAIWire.com.