Google is taking one of its most ambitious AI infrastructure experiments beyond Earth.

The company is preparing to launch a prototype satellite carrying its Tensor Processing Units (TPUs), the custom processors Google uses for artificial intelligence workloads, as part of its Project Suncatcher research programme.

The mission is designed to answer a fundamental question: can AI computing hardware operate reliably in space?

Google says the satellite will collect real-world data on how its TPUs handle the physical stresses of launch, radiation and the extreme thermal environment of low Earth orbit. The experiment is an important early step toward Google's longer-term idea of building scalable machine-learning infrastructure in space.

From AI Data Centres on Earth to Computing in Orbit

Project Suncatcher was first announced by Google in November 2025 as a research moonshot exploring whether interconnected, solar-powered satellites could eventually host large-scale machine-learning infrastructure.

The concept is based on a simple but ambitious proposition: put AI computing closer to a virtually continuous source of solar energy.

According to Google, satellites in low Earth orbit can receive significantly more sunlight than solar installations on Earth because they are above the atmosphere and can spend substantial periods of their orbit exposed to the Sun. Google estimates that satellites could generate up to eight times more solar power than equivalent solar infrastructure on Earth under the right conditions.

The company is not suggesting that conventional data centres are about to disappear.

Instead, Suncatcher is investigating whether space could eventually become another location for large-scale AI computing, particularly as demand for processing power continues to grow.

The First Test Is About Survival

Before Google can consider putting dozens or hundreds of AI processors into orbit, it has to establish that those processors can survive the journey and continue operating.

A rocket launch exposes electronics to intense vibration and acceleration.

Google says a trip into low Earth orbit lasts roughly 10 minutes, during which the spacecraft can experience loads of up to 10 times Earth's gravity. Individual components can experience considerably higher forces.

The company conducted ground-based vibration tests by shaking the satellite across multiple axes to replicate the conditions of launch.

Radiation is another major concern.

Unlike computers operating on Earth, satellites are exposed to radiation from solar activity and cosmic rays. These particles can interfere with electronic components and cause errors such as bit flips.

Google says it previously tested its Trillium TPUs in a proton-beam facility at the University of California, Davis, while the chips were running AI workloads. The company's initial results indicated that the TPUs could withstand a total ionising radiation dose greater than what they would encounter during a five-year space mission.

But laboratory testing has limits.

The upcoming orbital mission is intended to provide data from the environment Google ultimately wants its hardware to operate in.

Cooling Could Be One of the Biggest Problems

Keeping AI chips cool is already a major engineering challenge on Earth.

The problem becomes even more complicated in space.

On Earth, data centres can move heat through air or liquid cooling systems. In the vacuum of space, there is no surrounding air to carry heat away.

That means a space-based data centre would have to rely heavily on radiators and other specialised thermal systems to move heat away from processors.

Google says it is testing a combination of heat pipes and radiators for its Suncatcher hardware. The company has already tested the cooling technology inside thermal-vacuum chambers designed to reproduce the conditions of space.

The upcoming mission will show how those systems perform outside the laboratory.

That information could become just as important as the performance of the TPU itself.

A processor that can survive radiation but cannot efficiently dispose of the heat it generates would not be useful as part of a large orbital computing network.

Google Has a Bigger Network in Mind

The first satellite is only the beginning of the experiment.

Google's longer-term Suncatcher concept involves multiple satellites working together, with each spacecraft potentially carrying dozens of TPU chips.

The satellites would need to communicate with one another at extremely high speeds if they were going to function as a distributed AI computing system.

Google plans to use laser-based communications for this purpose.

That introduces another engineering challenge: satellites moving rapidly through orbit need to maintain extremely precise connections with their neighbours while simultaneously transferring huge volumes of data.

Google says it plans to test this aspect of the project in 2027, when two satellites are expected to be placed in orbit to evaluate the inter-satellite laser communication system.

Why Google Is Looking Beyond Traditional Data Centres

The experiment comes as AI computing requirements continue to expand.

Training and running increasingly sophisticated AI models requires enormous amounts of computing power, electricity, cooling infrastructure and physical space.

Google has invested heavily in its own AI hardware as a way to improve computing efficiency. Its latest eighth-generation TPUs, TPU 8t and TPU 8i, are designed specifically for large-scale training and inference workloads. Google says TPU 8t can scale to 9,600 chips in a single superpod and deliver 121 exaflops of compute.

At the same time, Google acknowledges that the growth of AI is putting new pressure on electricity infrastructure.

Its 2026 environmental report said Google's electricity demand increased 37% year over year in 2025, while the company continued investing in additional clean-energy capacity.

That makes the Suncatcher experiment particularly interesting.

If computing could eventually be moved into orbit and powered primarily by solar energy, space could offer a fundamentally different infrastructure model.

But that possibility remains highly experimental.

This Is Not a Space Data Centre Yet

It is important to distinguish between Google's current test and the much larger vision behind Project Suncatcher.

The first mission is essentially a technology demonstration.

Google needs to determine whether the processors, thermal systems, communications technologies and other components can reliably operate in orbit before a much larger deployment could even be considered.

There are also questions around launch costs, maintenance, radiation protection, communications bandwidth, orbital congestion, hardware replacement and the economics of operating computing infrastructure hundreds of kilometres above Earth.

A satellite constellation capable of supporting serious AI workloads would require much more than simply putting existing data-centre hardware on a rocket.

Google itself describes the project as a long-term research effort rather than a ready-to-deploy commercial data-centre platform.

The Bigger AI Infrastructure Race

Project Suncatcher reflects a broader shift in how technology companies are thinking about AI infrastructure.

The AI race is no longer only about who has the most capable model.

It increasingly involves who can secure enough chips, electricity, cooling, networking capacity and physical infrastructure to run those models at scale.

Google's experiment asks whether some of that infrastructure could eventually exist somewhere other than on Earth.

If the technology works, future AI systems could potentially be distributed across networks of solar-powered satellites, with processors communicating through high-speed optical links.

That scenario is still years away, and many technical and economic hurdles remain.

But the first step is considerably more basic: prove that the hardware can survive space.

That is what Google's upcoming mission is designed to find out.

Google's Suncatcher experiment is interesting because it attacks AI's infrastructure problem from an unusual direction.

Instead of asking only how to make processors faster or data centres more efficient, Google is asking whether the location of computing itself could change.

The experiment does not mean AI data centres are moving into space tomorrow. But if Google can demonstrate that its processors can survive radiation, manage heat and communicate reliably in orbit, it could establish some of the technical foundations for a completely different model of AI infrastructure.

For now, the satellite is a test. The bigger idea is the moonshot.