The Hardware Has Already Landed
The first hardware data centre is already on the Moon . Not in a conceptual white paper, not in a venture-capital pitch deck, but physically installed on lunar regolith, processing data 384,400 kilometres from the nearest Ethernet cable. That fact alone marks a threshold crossed, but it is merely the opening salvo in what has become a remarkably crowded—and consequential—race to move humanity's computational infrastructure into orbit.
In December 2025, a startup called Starcloud, backed by Y Combinator and Nvidia, sent its first AI-equipped satellite into space . By early 2027, Google will begin construction of AI data centres in orbit , whilst Aetherflux plans to launch its first data centre satellite as part of a constellation it calls 'Galactic Brain' . That same year, Axiom Space—bolstered by a $5.5 million grant from the Texas Space Commission—will develop its own orbital data centre in low-Earth orbit . Meanwhile, SpaceX has filed plans with the Federal Communications Commission for an orbital data centre constellation of up to one million satellites . Sam Altman is exploring building, funding or buying a rocket company to compete with Elon Musk in what has become, quite literally, a space race . Even the European Space Agency has entered the fray, awarding Edge Aerospace a contract to study the future of orbital data centres under its Space Cloud programme .
The speed and scale of these developments suggest something more profound than iterative progress. This is a structural shift in how—and where—the world's most powerful computational systems will be built. The question is no longer whether data centres will move into space, but what happens when they do.
The Physics of Going Up
The engineering logic is seductive. Project Suncatcher, for instance, envisions compact constellations of solar-powered satellites carrying Google's Tensor Processing Units, connected by free-space optical links for scalable AI compute in space . Penn engineers have developed a novel design for solar-powered data centres that will orbit the Earth , whilst Nvidia has sent its H100 GPU to space for the first time to test how data centres could work in orbit . These are not speculative proposals; they are hardware tests, institutional commitments, regulatory filings.
The advantages are material. In space, solar panels face the sun continuously, unobstructed by weather, night or atmosphere. On Earth, even the most efficient photovoltaic installations suffer from intermittency and degradation—the National Renewable Energy Laboratory estimates an annual efficiency loss of 0.5% to 0.8% due to ultraviolet exposure and adverse conditions . In orbit, that degradation curve flattens. Power is abundant, constant and—crucially—free of the carbon accounting that now haunts every terrestrial hyperscale facility.
Then there is heat. Data centres are, at their core, devices for converting electricity into computation and waste heat. On Earth, that heat must be managed with water, air conditioning and increasingly baroque cooling architectures. In the vacuum of space, radiative cooling becomes vastly more efficient. The thermodynamic problems that constrain ground-based facilities dissolve.
Google's chief executive, Sundar Pichai, has framed the shift in almost mundane terms. "A decade or so away, data centres in space will be viewed as a more normal way to build them," he said . The phrasing is telling. Not revolutionary. Not experimental. *Normal*. The implication is that the infrastructure now being tested is not a moonshot—forgive the pun—but the foundation of an industrial reordering already underway.
The Crowded Frontier
Yet the race is not occurring in a vacuum, literal or otherwise. Multiple companies and countries are rapidly increasing their launch capacity, with plans to deploy tens of thousands of satellites within a few years . Blue Origin, Jeff Bezos's aerospace venture, has announced plans to launch TeraWave, a satellite network with over 5,400 satellites, to compete with Starlink . The military is also staking its claim: the Space Development Agency has renamed its resilient layered network of military satellites the 'Proliferated Warfighter Space Architecture' , whilst the Pentagon is developing a Space Data Network as part of the Golden Dome missile-defence shield, designed to provide secure command and control to military leaders and enable rapid data transmission to interceptors .
The sheer density of what is planned is staggering. SpaceX's one-million-satellite constellation, if realised, would increase the number of objects in low-Earth orbit by orders of magnitude. The immediate competitors—Google, Blue Origin, Axiom, Starcloud, Aetherflux—are all targeting the same orbital bands, the same launch windows, the same constrained real estate. Earth orbit, as one analysis puts it, is "a limited resource with growing access, value and risk" .
That risk is not hypothetical. The spectre now haunting this gold rush is Kessler Syndrome: a fission-like runaway cascade of collisions in space. The risk of such an event is growing as the number of satellites increases exponentially . One collision produces debris. Debris strikes another satellite, producing more debris. The cascade accelerates. In the worst-case scenario, low-Earth orbit becomes unusable for generations, choked with shrapnel travelling at orbital velocities. It is a commons tragedy written in kinetic energy.
No binding international framework yet governs the deployment of these constellations. The Federal Communications Commission has jurisdiction over American actors, but its remit is spectrum allocation, not orbital safety. The outer-space treaty of 1967 established that space is the province of all mankind, but it predates the commercial satellite industry by decades and has no enforcement mechanism for congestion or collision. The gold rush, in other words, is proceeding ahead of the law.
The Computational Imperative
Why, then, the urgency? Why now?
The answer lies in the exponential growth of AI workloads. Training large language models, running inference at scale, processing real-time sensor data from millions of devices—these tasks are pushing terrestrial data centres to their operational and environmental limits. The electricity consumption of hyperscale facilities is now a geopolitical concern. Water use for cooling is coming under scrutiny in drought-prone regions. Land is finite. Regulatory opposition is mounting.
Space offers a release valve. Solar power is effectively infinite. Cooling is passive. There are no zoning boards, no neighbours to placate, no electrical grids to overload. The capital costs are enormous, but the operational economics—once the infrastructure is in place—are compelling. And the players driving this shift are not marginal actors. They are Google, Nvidia, SpaceX, Blue Origin, the Pentagon. Institutional capital is flowing in at scale.
"A decade or so away, data centres in space will be viewed as a more normal way to build them."
Pichai's timeline is worth parsing. A decade is not long in infrastructure terms. It suggests that the technical challenges—launch costs, radiation hardening, data transmission, orbital mechanics—are no longer seen as prohibitive. The engineering is within reach. What remains is deployment, iteration, scaling.
That confidence is bolstered by recent hardware tests. Starcloud's December 2025 satellite demonstrated that AI workloads can run in orbit . Nvidia's H100 test proved that cutting-edge GPUs can survive launch and function in vacuum . The lunar data centre, whatever its precise configuration, shows that the concept extends beyond low-Earth orbit . These are not thought experiments. They are operational proofs of concept.
The Military Dimension
But if the commercial logic is clear, the strategic logic is even sharper. The Pentagon's Space Data Network is not a research programme; it is a component of the Golden Dome missile-defence architecture, designed to provide secure command and control and enable rapid data transmission to interceptors . The Proliferated Warfighter Space Architecture is not a branding exercise; it is a networked constellation of military satellites designed to be resilient against attack .
The distinction between commercial and military infrastructure in space is already blurring. SpaceX's Starlink has been used extensively in the Ukraine conflict for battlefield communications. The same satellites that provide broadband to rural America also provide targeting data to artillery units. A data centre in orbit, ostensibly built to train large language models, could just as easily process reconnaissance imagery, coordinate drone swarms or relay encrypted commands to autonomous weapons systems.
This dual-use reality is not lost on the actors involved. Axiom Space is a private company, but it is building the next-generation commercial space station and receiving state funding for orbital data centres . Blue Origin is a commercial venture, but its heavy-lift rockets are being marketed to the Department of Defence. Google is a civilian corporation, but its cloud services are already integral to military and intelligence operations. The lines are not merely blurred; they are structurally entangled.
The implications are profound. If the next generation of computational infrastructure lives in orbit, then control of that infrastructure becomes a matter of national security. Launch capacity, spectrum allocation, orbital slots—all become strategic assets. The space race of the 1960s was driven by prestige and Cold War competition. This one is driven by the imperative to control the commanding heights of the digital economy and the military systems that depend on it.
The Environmental Ledger
The environmental calculus, meanwhile, is more ambiguous than the boosters suggest. On the one hand, moving energy-intensive computation off-planet could relieve pressure on terrestrial grids and water supplies. On the other, the launch industry itself is carbon-intensive. Rocket exhaust deposits black carbon and alumina directly into the stratosphere, where it persists and affects atmospheric chemistry in ways that are not yet fully understood.
Moreover, the commercial space race comes with multiple planetary health risks . The exponential increase in satellite numbers raises the probability of cascading collisions. Debris re-entering the atmosphere burns up, but the metals and composites release pollutants. The radio-frequency emissions from thousands of satellites interfere with radio astronomy, making it harder to study the universe. The night sky itself is changing; amateur astronomers report increasing interference from satellite trails, and indigenous communities have raised concerns about the cultural and spiritual significance of an increasingly crowded firmament.
These are not arguments against the technology per se, but they complicate the narrative that moving computation into space is a straightforward environmental win. The externalities are shifting, not disappearing. The question is whether the governance structures exist to manage them responsibly—and the answer, at present, is that they do not.
The Next Normal
So what does the next decade look like?
If the current trajectory holds, low-Earth orbit will host a rapidly expanding mesh of data centres, fibre-optic-equivalent laser links and solar arrays. Google's construction will begin in early 2027 . Aetherflux will launch its first Galactic Brain satellite around the same time . SpaceX will continue filing for ever-larger constellations . Blue Origin will deploy TeraWave . Axiom will build its orbital facility with Texas state funding . The European Space Agency will study, and likely deploy, its own systems . Starcloud and a dozen other startups will iterate, fail, merge or succeed.
The hardware is already proven. The capital is committed. The regulatory environment, such as it is, has not yet erected meaningful barriers. The military and commercial incentives are aligned. The engineering challenges—whilst non-trivial—are being solved in real time.
What remains uncertain is whether the governance will catch up. The Kessler Syndrome risk is real and growing . The dual-use nature of orbital infrastructure raises questions about accountability, sovereignty and the militarisation of space that the Outer Space Treaty does not adequately address. The environmental impact, both in orbit and in the atmosphere, is poorly understood and largely unregulated.
Sam Altman's exploration of building or buying a rocket company is emblematic. It signals that the AI industry no longer sees space as someone else's problem. It is infrastructure, and infrastructure must be owned, controlled and secured. The logic is the same as when Google began laying its own subsea cables or Amazon built its own logistics network. Vertical integration into the commanding heights.
Pichai's prediction that orbital data centres will become "normal" is probably correct . The question is what kind of normal we are building. A normal in which access to computation, like access to orbit, is concentrated in the hands of a few state and corporate actors? A normal in which the space around our planet becomes as congested, contested and commercially exploited as the ground beneath our feet? A normal in which the environmental and strategic risks are managed collectively, or one in which they are externalised until crisis forces a reckoning?
The first hardware data centre is already on the Moon . The rest is no longer speculative. It is happening, at scale, right now. The only question is whether the institutions, laws and norms that govern it will be built with the same speed and ambition as the satellites themselves.