Local opposition and resource scarcity are pushing tech giants to consider an unlikely solution for the growing demands of artificial intelligence: data centers in orbit.
In Tucson, Arizona, Councilmember Nikki Lee successfully opposed a proposed data center, Project Blue, citing concerns over water usage in the drought-stricken district. "We don’t think data centers belong in the desert," Lee stated, reflecting a sentiment that has become increasingly common.
This terrestrial resistance, coupled with the high costs and long wait times for power grid connections, is leading companies like Google, Meta, and SpaceX to investigate launching data centers into space. Amazon Web Services withdrew from the Tucson project after the city council's rejection, and Blue Origin has filed plans for thousands of potential orbital data centers.
While launching hyperscale AI data centers into orbit presents significant engineering hurdles, immense costs, and potential environmental risks, these challenges may be preferable for AI companies compared to facing populist backlash and infrastructure delays on Earth. SpaceX, for example, has publicly stated its intention to build an orbital AI data center by 2028, aiming to harvest 100 gigawatts of solar energy annually for AI operations. Google also released a study suggesting that orbital data centers could become cost-competitive with terrestrial ones within a decade as launch costs decrease.
Experts agree that the technical feasibility of space-based data centers is no longer the primary question, but rather the timeline for their readiness and the economic viability of powering large-scale AI operations from orbit. Currently, building a one-gigawatt data center in space is estimated to cost around $51.5 billion, compared to $16 billion on Earth, according to aerospace engineer Andrew McCalip. Jason Aspiotis, a space systems engineer, acknowledged the engineering challenges as solvable but pointed to the economics as the main uncertainty.
Even if space-based data centers become a reality, their impact on Earth remains uncertain. Critics argue that such ventures would not solve Earth's infrastructure problems like affordable housing or public transit and could divert crucial resources and tax revenue into space. Some politicians, like Virginia Delegate David Reid, who represents Loudoun County, a hub for data centers, dismiss the idea of orbital data centers as "fictional nonsense" and a distraction from local land-use planning.
Beyond the economic and engineering aspects, significant practical concerns exist. An orbital data center would require extensive cooling systems, potentially larger than football fields, and would need to be assembled in space, likely by robots. The constant solar radiation could damage sensitive circuitry, and repairs would be impossible to conduct by sending up a technician. Furthermore, the proliferation of space junk from these facilities could threaten existing orbital infrastructure, such as GPS and weather satellites. Sven Bilén, a professor of aerospace engineering, suggests that early orbital data centers will likely focus on processing data from space, like satellite imagery, rather than consumer-facing AI applications.
However, the significant financial backing from companies like SpaceX and Google, coupled with the increasing difficulty of building data centers on Earth, might accelerate the development of orbital compute. Krishna Muralidharan from the University of Arizona predicts that orbital data centers will exist in some form within five years, though the business case remains to be seen.
The potential for millions of new satellites to support space-based AI could also drastically alter the night sky. Astronomers and environmental advocates express concern that the increase in artificial satellites could outnumber visible stars, changing humanity's relationship with the cosmos. This concern mirrors the debate surrounding terrestrial data centers, which often face local opposition due to their scale and impact on communities.