As artificial intelligence workloads strain the limits of terrestrial infrastructure, a bold and controversial idea has gained traction among tech entrepreneurs and aerospace engineers: launching data centers into orbit. The concept, which sounds ripped from science fiction, is being pursued by real companies with real funding. But a growing chorus of scientists and environmental researchers warns that the cure could prove worse than the disease, potentially accelerating the very climate problems these orbital facilities are designed to sidestep.
The appeal is straightforward. AI model training and inference require enormous amounts of electricity, and the demand is growing at a pace that has alarmed utility companies and grid operators across the United States and Europe. Data centers already consume roughly 1.5% of global electricity, and that figure is projected to rise sharply as generative AI applications proliferate. Cooling these facilities — which generate tremendous heat — adds another layer of cost and environmental burden. In space, the thinking goes, solar energy is abundant and uninterrupted, and the vacuum provides a natural heat sink.
The Companies Betting on Orbital Computing
Several startups have emerged in recent years with plans to put computing hardware above the atmosphere. Lumen Orbit, a Y Combinator-backed company, has been among the most visible, proposing satellite-based data centers powered by solar arrays. Other ventures, including some backed by defense-adjacent investors, have floated similar concepts. The pitch to investors typically centers on three advantages: virtually unlimited solar power with no night cycle (in certain orbits), passive cooling via radiative heat dissipation into space, and freedom from the land-use and permitting constraints that slow terrestrial data center construction.
As Engadget reported, the technical feasibility of orbital data centers is not entirely in question — the physics of solar power and radiative cooling in space are well understood. What is deeply uncertain, and increasingly alarming to researchers, is the environmental toll of building, launching, and eventually deorbiting the thousands of satellites such a system would require.
The Launch Problem: Rockets, Soot, and the Upper Atmosphere
The most immediate environmental concern involves the rockets themselves. Every launch deposits black carbon — soot — directly into the stratosphere, where it is far more potent as a warming agent than the same material released at ground level. According to research cited by Engadget, black carbon particles in the stratosphere can persist for years and are estimated to be roughly 500 times more effective at trapping heat than equivalent emissions at the surface. Unlike carbon dioxide from power plants, which disperses through the lower atmosphere, stratospheric soot sits in a region where it can directly absorb solar radiation and alter atmospheric chemistry.
The current generation of heavy-lift rockets, including SpaceX’s Falcon 9 and the forthcoming Starship, burn kerosene-based or methane-based fuels that produce significant quantities of this soot. A single Falcon 9 launch deposits an estimated 50 tons of carbon dioxide equivalent into the upper atmosphere. Scaling orbital data center operations to a level that would meaningfully offset terrestrial energy consumption would require launch cadences far beyond what the industry currently sustains — potentially hundreds or thousands of additional flights per year.
Ozone Depletion and the Reentry Debris Question
Beyond soot, there is the question of ozone depletion. Rocket exhaust contains nitrogen oxides and alumina particles that catalyze the destruction of ozone molecules in the stratosphere. The ozone layer, which shields the Earth’s surface from harmful ultraviolet radiation, has only recently begun to recover from decades of damage caused by chlorofluorocarbons. Researchers have warned that a dramatic increase in launch frequency could reverse some of those gains. A 2022 study published in the journal Earth’s Future found that projected increases in rocket launches could cause measurable ozone loss, particularly over polar regions.
Then there is the problem of what happens when orbital data centers reach end of life. Satellites in low Earth orbit eventually deorbit, burning up in the atmosphere upon reentry. This process releases metallic particles — aluminum, lithium, copper, and other materials used in electronics and solar panels — into the upper atmosphere. The long-term effects of depositing large quantities of vaporized metals into the stratosphere are not well understood, but early research suggests they could also contribute to ozone depletion and alter the reflective properties of the atmosphere in unpredictable ways. As Engadget noted, the sheer volume of hardware required for orbital data centers would dramatically increase the mass of material reentering the atmosphere each year.
The Energy Math Doesn’t Quite Add Up
Proponents argue that the environmental costs of launches would be offset by the elimination of fossil-fuel-powered electricity generation on the ground. But this argument rests on assumptions that many energy analysts find questionable. For one, terrestrial data centers are increasingly powered by renewable energy. Microsoft, Google, and Amazon have all signed massive power purchase agreements for wind, solar, and nuclear energy. The marginal emissions reduction from moving computing to orbit diminishes as the terrestrial grid gets cleaner.
Moreover, the energy density of computing hardware that can survive launch vibrations, radiation exposure, and the thermal cycling of orbit is significantly lower than what can be achieved in a climate-controlled building on the ground. Radiation hardening alone — the process of making chips resistant to cosmic rays and solar particle events — imposes performance penalties that could negate much of the efficiency gained from free solar power and passive cooling. Maintaining and upgrading hardware in orbit is also orders of magnitude more expensive and logistically complex than swapping out server racks in a warehouse in Iowa.
Latency, Bandwidth, and the Practical Limits of Space Computing
There are also fundamental networking constraints. Data must travel between orbital servers and terrestrial users, introducing latency that varies depending on orbital altitude. Low Earth orbit satellites at roughly 550 kilometers altitude impose round-trip latencies of approximately 20 to 40 milliseconds — acceptable for some applications but problematic for real-time AI inference, financial trading, and other latency-sensitive workloads. Bandwidth is another bottleneck; optical and radio links between ground stations and satellites have finite capacity, and weather can degrade signal quality.
For AI model training, which involves moving enormous datasets and requires high-bandwidth interconnects between thousands of processors, the constraints are even more severe. Modern training clusters rely on ultra-low-latency, high-bandwidth networking fabrics like NVIDIA’s NVLink and InfiniBand. Replicating that connectivity between satellites — each subject to orbital mechanics, varying distances, and signal propagation delays — presents engineering challenges that no company has yet demonstrated a credible solution for at scale.
Regulatory Uncertainty and the Debris Risk
The regulatory environment adds another layer of complexity. Space is governed by a patchwork of international treaties and national regulations that were not designed for commercial-scale orbital infrastructure. The Outer Space Treaty of 1967 holds nations responsible for objects they launch, but enforcement mechanisms are weak. The growing problem of space debris — there are already tens of thousands of tracked objects in orbit, and millions of smaller fragments — raises the risk of collisions that could generate cascading debris fields, a scenario known as Kessler Syndrome. Adding thousands of large data center satellites to an already congested orbital environment would significantly increase collision risk.
The Federal Communications Commission and the Federal Aviation Administration both have roles in regulating satellite launches and orbital operations, but their mandates do not extend to comprehensive environmental review of stratospheric emissions from rockets. The Environmental Protection Agency has historically not regulated rocket launches under the Clean Air Act. This regulatory gap means that the atmospheric effects of a massive scale-up in launch activity could go largely unmonitored and unmitigated.
A Familiar Pattern: Solving One Problem by Creating Another
The orbital data center concept fits a recurring pattern in technology: addressing a pressing problem by externalizing costs to a different domain. Terrestrial data centers consume too much electricity and water, so the proposed solution shifts the burden to the upper atmosphere — a commons with even less regulatory protection than groundwater or the electrical grid. The stratosphere has no utility company to send a bill, no local planning commission to deny a permit, and no neighbors to file complaints about noise or heat.
None of this means that space-based computing is inherently a dead end. There may be niche applications — such as Earth observation data processing, or computing for space-based operations themselves — where orbital hardware makes practical and environmental sense. But the vision of orbital data centers as a wholesale replacement for terrestrial AI infrastructure appears, on current evidence, to trade a set of manageable and improving terrestrial problems for a set of poorly understood and potentially irreversible atmospheric ones.
The AI industry’s hunger for compute is real and growing. But the answer to that hunger is more likely to be found in next-generation nuclear reactors, more efficient chip architectures, and smarter software — not in filling the sky with server farms whose environmental costs we are only beginning to comprehend.