AI data centres in space face scientific pushback
The promise sounds like science fiction: move the massive computer warehouses that power artificial intelligence into orbit, where the sun always shines and no community can object. . SpaceX, the rocket company led by Elon Musk, is among a handful of firms planning to launch constellations of satellites that would act as data centres in low-Earth orbit. . The logic appears seductive, almost inevitable. Abundant solar energy, no land-use disputes, no angry neighbours holding protest signs. . But this vision rests on a flawed premise, one that scientists are now pushing back against with a very different proposal. .
The argument from space enthusiasts is that data centres are simply a necessary cost of progress. . If AI is to drive scientific breakthroughs, so the thinking goes, it must be powered by enormous facilities consuming staggering amounts of electricity and water. . Yet this equation ignores a crucial distinction. The infrastructure needed for scientific research is fundamentally different from the infrastructure needed to run consumer AI platforms serving millions of everyday users. . A chatbot answering questions for the public is not the same as a researcher analysing a specific dataset. They do not require the same scale, the same energy, or the same approach. .
A Smaller Footprint, A Bigger Voice
The alternative vision comes from researchers who understand both the science and the technology. . Cassidy K. Buhler, a postdoctoral fellow at the Cooperative Institute for Research in Environmental Sciences and the Environmental Data Science Innovation and Impact Lab at the University of Colorado Boulder, is one of the voices championing a different path. [2] Alongside Fernando Perez and Carl Boettiger, both affiliated with the Eric and Wendy Schmidt Center for Data Science and Environment at the University of California, Berkeley, they argue for a shift away from mega data centres entirely. [3] Their proposal centres on open-weight AI models — systems whose parameters are publicly available for anyone to inspect, modify, and deploy. .
Open-weight models change the fundamental relationship between researcher and tool. . Instead of sending data to a distant corporate server, scientists can run the model locally, on their own hardware, with their own data. . This is not a marginal technical detail. It transforms who controls the AI, where the computation happens, and how much energy the process consumes. . The researchers argue that this approach is more sustainable, more aligned with public interest, and gives scientists far greater control over the tools they use. . For a community increasingly concerned about the environmental cost of computation, this is a meaningful shift in direction. .
Public opposition to data centres is already a visible force. In San Marcos, Texas, activists have protested against proposed facilities that would power artificial-intelligence systems. . The image of residents holding signs against a concrete building filled with servers is becoming familiar across the globe. . These protests reflect a deeper anxiety about who bears the cost of technological progress — and who benefits from it. . When communities resist data centres, they are not opposing science; they are opposing a model of development that demands their resources without their consent. .
The Real Cost of Computation
The energy question is not abstract. It is measurable, and it is growing. . The researchers point to the soaring demands of data centres as a primary driver of public concern. [2] Water is another flashpoint. Data centres require enormous amounts of cooling, which in many regions means consuming precious freshwater supplies. . When technology companies propose moving these facilities to space, they are attempting to sidestep these terrestrial constraints. . But the researchers argue that this is a solution to a problem that should not exist in the first place — because the problem is not AI itself, but the scale at which it is being deployed. .
How much energy will AI really consume? That question remains open, with estimates ranging widely depending on assumptions about growth and efficiency. . What is clear is that the current trajectory is unsustainable, and that the scientific community has both the knowledge and the responsibility to chart a different course. . The researchers are not calling for an end to AI in science. They are calling for a rethinking of how AI is delivered — locally, efficiently, and with public oversight rather than in distant, resource-hungry facilities that provoke resistance wherever they are proposed. .
The limitation of this approach is also its strength. Open-weight models require technical expertise to deploy effectively, and not every research group has that capacity. . The researchers acknowledge that scientists with the necessary know-how must champion this vision, implying that the burden falls on those who can to lead the way for those who cannot. . This is not a universal solution, but a targeted one — appropriate for the scientific community, which values transparency, reproducibility, and control. . Whether the broader AI industry follows remains uncertain, but the scientists’ argument stands on its own merits. .
Sources
2. Cooperative Institute for Research in Environmental Sciences