AI gets into the space race: technology companies’ plan to take data centers off Earth

As AI ambitions expand, so does interest in moving some computing architecture beyond the surface. Governments, startups and investors are already evaluating this opportunity.

The initiative is far from being an immediate solution. However, as power grids become saturated and new projects run into new obstacles, space is beginning to appear less as an extravagance and more as an alternative that deserves serious evaluation.

The advantage of this deployment is beyond the total immunity against floods, earthquakes or local conflicts. The orbital vacuum allows us to operate far from the regulatory, fiscal and environmental obstacles that currently slow down the growth of traditional data centers. Basically, space offers the unlimited scope for expansion that the Earth’s surface began to restrict.

The space race takes on a new protagonist. It is no longer just about launching rockets or deploying satellites: Google, Amazon, Microsoft, SpaceX y xAI They explore how to transform Earth’s orbit into a gigantic platform for processing data.

Starcloud He is among the pioneers of this new frontier. The startup imagines a constellation of up to 88,000 satellites functioning as a data center distributed around the planet. Instead of concentrating thousands of servers in monolithic buildings, the proposal seeks to distribute processing between spatial nodes connected by laser links and powered by solar energy.

Orbitalfounded by the young Euwyn Poon, is moving forward with a similar approach based on data centers powered by continuous solar energy. Its goal is to build processing platforms in low Earth orbit capable of harnessing the energy available outside the atmosphere and dissipating heat through radiation. The company plans to carry out its first operational tests during 2027.

SpaceXsupported by the experience accumulated with Starlink, studies how to expand its future constellations. The goal would not be to transmit information, but also process it in space. If realized, the orbital network could evolve into a gigantic cloud distributed around the planet, capable of executing workloads without depending on terrestrial data centers.

Through the Suncatcher Project, Google explores how to move its computing power into space. To this end, it partnered with Planet, one of the largest private Earth observation satellite operators, to launch two prototypes in 2027. Equipped with TPU chips and high-speed laser links, they will be used to audit border computing capacity (edge computing) in the orbital vacuum.

The ultimate goal is to create a distributed processing platform capable of harnessing a virtually inexhaustible energy source and scaling far beyond the limitations faced by terrestrial infrastructure today. The first tests already include resistance tests of the chips against space radiation and optical communication systems between satellites.

Launching a nanosatellite is already a complex task. Building a data center in orbit is a challenge of another magnitude. Each kilogram transported increases costs and requires components capable of withstanding extreme temperatures, micrometeorite impacts and years of exposure to radiation that can alter the operation of the chips.

Furthermore, the margins of error are reduced to a minimum. What can be solved on the ground during a maintenance visit, in orbit requires extreme levels of reliability, since any subsequent intervention is slow, delicate and exceptional.

Managing the temperature of orbital servers represents a complex financial and logistical challenge. Although the low outside temperature suggests simple cooling, the absence of matter cancels conventional thermal dissipation by conduction or convection. Thus, a theoretical advantage of the cosmos becomes its main operational bottleneck.

This limitation requires the incorporation of voluminous thermal radiation systems that add weight, complexity and costs launch. In other words, a significant part of the infrastructure ends up destined not to process data, but to keep the equipment that processes it operational.

Each additional measure forces us to sacrifice part of the available capacity for other functions. This redundancy is key to compensate for years of exposure to radiation and energetic particles that degrade critical components. As often happens in space engineering, much of the effort ends up focused on guaranteeing the survival of the system rather than maximizing its performance.

Additionally, each new constellation incorporates thousands of additional objects around the planet. The risk of collisions increases and each potential impact can generate fragments that remain in orbit for years. Specialists warn that space proliferation requires increasingly sophisticated tracking systems to avoid an orbital environment that is difficult to manage.

Research published by IEEE Spectrum concludes that operating a GPU in space is still several times more expensive than doing so in a terrestrial data center. The energy savings are not enough to offset the costs associated with launch, shielding, maintenance and hardware replacement. Physics, for now, continues to impose conditions that are difficult to negotiate.

In other cases orbital processing can offer specific advantages. Such as analysis of satellite images in orbit, space surveillance, early threat detection and satellite traffic management. In these cases, processing data close to its source reduces the need to transmit huge volumes of information back to Earth and improves response times.

Space data centers are no longer the realm of science fiction. Nor do they represent an inevitable revolution. The technology exists, business interest grows and investments begin to multiply. But between the promise of unlimited energy and the reality of thermodynamics there appears a considerable distance.

The future probably does not depend on replacing terrestrial data centers, but on finding those tasks where orbit contributes a genuine advantage. The question is no longer whether they can be built. The real question is determining when, where and why it is really worth doing it.