With demand for data centers seemingly growing by the day, tech businesses are racing to find solutions to the high costs and logistical difficulties of building the infrastructure needed to power cutting-edge AI. One novel proposal that has been advanced is to build data centers in space as a means of solving some of their most consistent challenges. Will there be data centers in space in the next few years, or is the idea of launching large amounts of computing infrastructure into orbit still too costly to make economic sense?
Advantages of Space-Based Data Centers
Although they may seem like science fiction, there’s a real case to be made for putting data centers in space. To begin with, space-based data centers address very real concerns about the availability of land and suitable local infrastructure, both of which can be problems for businesses trying to build large new data centers. Tied closely to these factors is the growing pushback against data center construction at the local level, which has blocked or delayed an estimated $64 billion worth of data center projects to date.
On the technical side, the appeal of space as a setting for data centers becomes even more significant. By making direct use of solar energy as a power source, space-based data centers could solve one of the largest constraints on terrestrial computing power, namely a growing shortage of power generation capacity. Removing the need to purchase or generate power externally would also cut out a massive ongoing cost incurred by traditional data centers.
According to Starcloud, a startup involved in exploring the potential for large-scale data centers in space, the energy costs could be as little as one-tenth of the cost for ordinary data centers, even accounting for the large initial expense of launching hardware into space. Though such projections seem appealing, there are a few challenges that could mitigate the seemingly strong appeal of space-based data centers.
Existing Economic Challenges of Terrestrial Data Centers
While there are some compelling arguments for building space-based data centers, it may also be helpful to bring things back down to Earth by considering the difficulties even ordinary data centers are facing. Specifically, data centers being built today suffer from extremely rapid depreciation that makes it difficult for them to achieve profitability.
Whether in space or on Earth, it’s also worth taking into account the short lifespans imposed on data centers by the speed at which hardware becomes obsolete. AI accelerators have a roughly 5-year hardware lifespan, and chip manufacturers are currently making great strides in computing power with each passing year. As a practical matter, this means that AI infrastructure built today becomes obsolete compared to the most cutting-edge technology within a very few years.
These factors even played into a recent warning from IBM CEO Arvind Krishna about the challenges of profitably building data centers to power AI. Krishna, citing hardware costs and depreciation, called into question whether the massive capital expenditures that currently define the tech world’s approach to AI were economical. Despite advantages in maintenance and power consumption, these factors would also seem to work against space-based data centers. Replacing the hardware would also likely be massively more expensive due to the cost associated with launching new hardware into orbit.
Can Data Centers Be Built in Space at Scale?
Another potential sticking point when it comes to building AI infrastructure in space is scale. On paper, space-based data centers are actually more scalable than their terrestrial counterparts due to the lack of constraints on available land. On the other side of that coin, however, is the high cost of launching GPU hardware into space, which could make data centers prohibitively expensive to build in space at scale.
Despite claims of superior economics by the likes of Starcloud, launch costs remain a serious sticking point. To launch a single kilogram of material into space currently costs well over $1,000, making the building of large-scale data centers in orbit unfeasible for the time being. Although businesses that are already building satellite constellations could have an advantage, the cost to build something comparable to a large data center campus on Earth would likely still be prohibitive.
It’s also worth noting that GPUs operating in space need specialized cooling infrastructure that must also be put in place at similar launch costs. Because air can’t be used to dissipate heat in the vacuum of space, processors operating in that environment require radiator cooling.
Finally, building in space introduces risks that aren’t present on Earth. To begin with, objects in orbit are subject to cosmic rays and other forms of radiation that introduce the need for radiation-resistant hardware. Security is also a very real concern, as space infrastructure is generally considered vulnerable to a host of cybersecurity risks that can be difficult to address due to lack of easy access.
Will There Be Data Centers in Space?
Right now, enough tech firms and startups are exploring the potential for space-based data centers that at least some computing infrastructure in space appears inevitable. These projects, however, are likely to remain at very small scales for the foreseeable future, providing proof of concept but likely not contributing meaningful computing power.
In the long run, it’s possible that enough of the engineering challenges associated with putting data centers in orbit could be solved to make it economically feasible. Certainly, the constraints of land and power associated with building ordinary data centers create a strong set of incentives for businesses working toward space-based data centers.
For now, however, it seems unlikely that large-scale data centers will be built in space within the next several years. Given the high costs of launching material into space, the other engineering problems that still need to be solved and the seemingly shaky economics of even terrestrial AI data centers, it will likely be quite some time before data centers in space become a meaningful part of the world’s AI infrastructure.
The author has no position in any of the stocks mentioned. Financhill has a disclosure policy. This post may contain affiliate links or links from our sponsors.