What happens when digital systems outgrow the Earth that sustains them? Computing has always consumed energy and water, but the growing ambitions of the AI industry are pushing demand to an unprecedented scale. As pressures mount, data centers in orbit around the Earth have been pitched as a solution. But space is no escape, and what is framed as a technical fix risks entrenching digital dependence in an increasingly militarized, fragile, and crisis-prone environment. Orbital data centers simply relocate critical infrastructure into an already contested domain where the boundaries between civilian and military systems blur, satellites become strategic targets, and systemic disruption becomes a tool of coercion.
Water as a Hidden Driver of Digital Infrastructure Conflict
At the core of this issue is the intersection of water, energy, and computing power. Artificial Intelligence (AI) infrastructure depends on already strained freshwater and energy systems for cooling, energy generation, and semiconductor manufacturing. Research by Li et al. estimates that training large language models such as GPT-3 may be associated with the consumption of as much as 700,000 liters of freshwater, depending on data center location and cooling practices. Semiconductor fabrication for AI chip production can also be highly water-intensive, with advanced facilities reportedly using up to 10 million gallons of ultrapure water (hundreds of times cleaner than drinking water) daily. Importantly, these demands do not arise from a fundamentally new category of infrastructure, but from the scaling of pre-existing computational systems that have long depended on water- and energy-intensive processes.
This growing demand amplifies pre-existing spatial inequalities related to the location of infrastructure, the concentration of data centers in regions already experiencing water stress, and the subsequent translation of local hydrological constraints into systemic pressures on digital infrastructure (see studies by Kseibati and Siddik). In line with this trend, Zohar Barnett-Itzhaki reports that approximately two-thirds of data centers built after 2022 are located in water-stressed areas.
Another fundamental constraint at the nexus of cooling, energy, and semiconductor performance is: the physics of heat transfer. AI data centers produce vast amounts of heat that must be managed. It is at this point in the debate—where thermal limits and resource availability begin to converge under increasing computational load—that we begin to find proposals to build orbital data centers. Within the cold vacuum of space, and alongside the continuous flux of the Sun, orbital environments are sometimes framed as offering relief from terrestrial water and energy constraints. Yet this intuition overlooks a key problem, namely that the orbital and atmospheric environment that is imagined as a passive receptacle for the undesirable impacts of AI infrastructure impacts would also be the medium on which that infrastructure would depend. The proposed solution thus amounts to an “exo-atmospheric” externalization of infrastructure costs, in which inequality is not resolved but re-inscribed into a system where valuable positions in orbit are effectively reserved for those with the resources to access and sustain them.
Solving Earth’s Problems by Exporting Them
Proposals for orbital data centers rest on the all-too-common assumption that the constraints of a system can be overcome by externalizing problems into the environment. However, at sufficient infrastructural scale this assumption begins to break down. In the context of large-scale AI infrastructure, the same logic motivates proposals to relocate computation into orbital environments.
Above the Kármán Line: Orbital Dynamics
Yet the vacuum of space does not provide effortless cooling but relies on radiative heat transfer, a process requiring large surface areas and specific orbital configurations. Orbital geometries, including sun-synchronous and terminator orbits, need to be carefully selected so that spacecraft can maintain continuous solar illumination for power generation while preserving stable thermal orientation toward cold space. These demands compound the need for large and complex orbital systems to be embedded within already congested orbital domains and ultimately drive-up collision risk.
At the same time, running advanced AI workloads depends on specialized semiconductor hardware that is already constrained by high power density and tightly managed thermal budgets. Beyond thermal limitations, the vacuum of space introduces a persistent risk of single event upsets necessitating radiation hardening.
These systems are not only constrained in orbit but also depend on repeated access to Earth’s atmosphere, linking orbital computing infrastructure directly back to launch dynamics and atmospheric impact.
Kármán Line
The Kármán Line is the commonly recognized boundary between Earth’s atmosphere and outer space, located about 100 km above sea level, where staying aloft requires orbital speed rather than conventional flight.
Below the Kármán line: Atmospheric physics
At the scale required for sustained orbital deployment of large computational infrastructure, rocket launches cease to be isolated events and instead become a persistent source of atmospheric perturbation.
Recent studies of black carbon emissions from rocket fuels (see studies by Maloney and Ross) highlight their role in modifying the radiative balance of the stratosphere. Absorption of solar radiation by black carbon leads to localized heating and perturbations in atmospheric chemistry.
A heavy launcher, carrying about 1,000 tons of kerosene fuel and using high-soot-emission engines, releases roughly 10 tons of black carbon into the stratosphere each time it launches. Black carbon absorbs solar radiation, leading to stratospheric warming. Increased temperatures can affect the rates of ozone-depleting chemical reactions. Aerosol emissions generated during atmospheric re-entry will also need to be considered if launches increase significantly, for example under scenarios involving satellite mega-constellations and orbital AI computing infrastructure, as described in recent SpaceX fillings and related reporting’s.
Maloney and colleagues conclude that if rocket engine technology remains unchanged and orbital launch frequency increases by an order of magnitude, significant stratospheric climate and ozone responses are expected, illustrating how expansion of infrastructure to the orbital scale does not decouple it from Earth system constraints.
Space Infrastructure as a Strategic Chokepoint
Space assets are inherently exposed to threats ranging from cyber-attacks and signal interference to anti-satellite weapons and orbital debris. As critical digital services become increasingly dependent on space-based systems, these assets may emerge as strategic targets in times of conflict. This creates a structural tension, in which infrastructure designed to enhance resilience by externalizing Earth-based constraints may also introduce new points of systemic vulnerability.
These challenges are compounded by limited international governance frameworks. Control over launch capabilities and orbital infrastructure remains a source of geopolitical leverage, contributing to asymmetries between states with access to space and those without. Questions surrounding ownership, resource allocation, and the militarization of dual-use technologies remain unresolved.
Proposals for satellite mega-constellations, including emerging concepts for space-based AI computing infrastructure, have raised significant concern within the astronomical community. Such systems may increase optical and radio-frequency interference, complicating ground-based observations that rely on long, uninterrupted exposures to detect faint astronomical signals and perform precise measurements. The astronomical community has engaged in ongoing efforts with industry to develop mitigation strategies to reduce these impacts. However, the effectiveness of such measures at very large scales remains uncertain. Some analysts suggest that extensive mega-constellations could substantially increase the number of visible satellites and contribute to elevated night sky brightness, thereby affecting observational astronomy globally.
In this context, orbital AI systems may also have implications for international security, particularly as space infrastructure becomes increasingly integrated into critical global systems.
Conclusion
Proposals to shift data centers into orbit fail to recognise that the true constraint may be neither launch costs nor radio spectrum allocation, but the stability of the upper atmosphere as an operating environment. Embedding critical digital infrastructure in an already congested and contested domain presents the strategic risks of systemic disruption, coercive regimes, and technical failures cascading across civilian and military systems. The challenge for policymakers is to align planetary stewardship with international security, thus ensuring sustainable AI development without hard-wiring new pathways for crisis and escalation in space.
