Project Suncatcher: What Google plans to test by taking AI compute to space

Google's Project Suncatcher will test AI chips in orbit as the company explores solar-powered satellite clusters as a potential new way to scale AI computing infrastructure

Google is looking beyond Earth for the next generation of artificial intelligence (AI) infrastructure. The company is preparing to launch the first test satellite for its Project Suncatcher on October 1, exploring whether AI computing can eventually be powered by solar energy in space. The prototype will carry four Google Tensor Processing Units (TPUs) to test how the chips perform under the radiation, thermal and other conditions of orbit.

 

According to Google, the mission is an early test rather than a plan to replace terrestrial data centres with satellites. But it addresses a growing challenge for the AI industry: finding enough power and infrastructure to support increasingly demanding computing workloads. Google's longer-term proposal involves clusters of satellites carrying AI processors, connected through high-speed optical links and powered by sunlight.

 

 

Google first announced Project Suncatcher in November 2025 as a research project exploring whether machine-learning infrastructure could eventually be scaled in space.

 

What is Project Suncatcher

 

Project Suncatcher is Google's research effort to explore whether AI computing can be performed in orbit instead of entirely inside data centres on Earth.

 

The basic idea is to put computing hardware, solar panels and cooling systems on satellites, use sunlight to generate electricity, process AI workloads on the satellites and connect multiple satellites so that they can operate as a larger computing system.

 

Google says satellites in low Earth orbit can receive near-constant sunlight and could generate up to eight times more solar power than comparable solar panels on Earth. That is one of the main reasons the company is studying space as a possible location for future AI infrastructure.

 

But Google is not saying that orbital data centres are ready to replace terrestrial ones. The company describes Suncatcher as a "moonshot" and says the first mission is intended to identify what works and what does not. The October launch is a test of the underlying technology, not the start of a commercial satellite data-centre network.

 

How do we put machine learning in space

 

The first requirement is computing hardware that can survive launch and then operate in orbit.

 

Google's AI processors are called Tensor Processing Units (TPUs). These are specialised chips designed to run machine-learning workloads. For Suncatcher, Google has been testing its Trillium TPUs to understand whether they can withstand the physical conditions of space.

 

Google says a launch into low Earth orbit lasts roughly 10 minutes, during which a spacecraft experiences strong vibration and acceleration. Individual components can experience forces of 50 to 100 times the force of gravity during launch.

 

Radiation is another concern noted by Google. According to the US-based technology giant, outside Earth's atmosphere, electronics are exposed to solar events and cosmic rays that can cause errors in computer hardware. Google says it tested its TPUs at the University of California, Davis' Crocker Nuclear Laboratory using a proton beam while running AI workloads.

 

The company said its initial tests showed the Trillium TPUs could withstand a radiation dose greater than what they would receive during a five-year space mission. The actual orbital test is intended to provide information that cannot be obtained in a laboratory.

 

For the October mission, the satellite is being developed in partnership with Planet and will fly on SpaceX's Transporter-18 rideshare mission. Google had earlier planned to launch two prototype satellites by early 2027, but the company is now conducting this earlier test to collect in-orbit data.

 

How much power can solar panels generate in orbit

 

A conventional data centre has to draw electricity from an energy grid or dedicated power infrastructure. Large AI data centres can require substantial amounts of electricity because thousands of processors may operate simultaneously.

 

A satellite in orbit has a different source of energy: the Sun. Google says solar panels in low Earth orbit can access near-constant sunlight and potentially generate up to eight times more solar power than on Earth. This does not mean every satellite automatically produces eight times as much electricity. The actual output depends on the size and efficiency of the solar array, the satellite's orientation and how much power the onboard systems consume.

 

For the first Suncatcher test, the objective is much smaller than building a full-scale orbital data centre.

 

For Google, the attraction is the possibility of combining abundant sunlight with increasingly efficient AI processors in an environment where large amounts of computing infrastructure could eventually be deployed without building another terrestrial power plant or data-centre campus.

 

The biggest problem may be cooling

 

Powering an AI processor is only half the problem. The heat produced by the processor also has to go somewhere.

 

On Earth, data centres use fans, air conditioning, chilled water and other cooling systems to move heat away from servers. Space works differently because there is no air around the satellite.

 

Google says this makes cooling one of the central engineering challenges for Suncatcher. In a vacuum, heat cannot simply be carried away by moving air. Instead, the satellite has to transfer heat to radiators, which then release that heat into space.

 

Google says it is testing a combination of heat pipes and radiators. It has also tested the cooling system inside a thermal vacuum chamber designed to replicate the conditions of space. The first satellite is therefore also a test of whether a small system can keep high-performance AI chips within their operating temperature range. 

 

How will TPU satellites communicate

 

A large AI data centre is not simply a collection of processors. The chips have to communicate rapidly with one another so that they can work together on the same workloads.

 

The same principle would apply if the processors were distributed across satellites.

 

Google said its long-term Suncatcher design involves groups of satellites carrying dozens of TPUs each. The satellites would need to exchange large amounts of data while moving around Earth.

 

Google plans to use laser-based communication links for this. The company says the technology already exists in space, but the requirements for Suncatcher are different. Existing systems often focus on communicating over long distances at lower bandwidth. Google's proposed system would need very high bandwidth between satellites that are relatively close to one another.

 

There is also a precision problem. As the satellites are constantly moving, their laser links have to remain accurately pointed at one another.

 

Google plans to test this part of the system in 2027, when it expects to put two satellites into orbit and study their communication capabilities.

 

What is the future of space-based compute

 

Google's longer-term plan is to build clusters of satellites carrying dozens of TPUs, with multiple satellites working together as a distributed AI computing system in orbit. The company says this could eventually provide another way to scale AI infrastructure.

 

But several questions remain, particularly around cost, cooling, radiation protection and communications. Launching and maintaining satellites is expensive, while some AI workloads may still need the low latency offered by data centres on Earth.

 

For now, Suncatcher remains a research project. Google plans to test two satellites and their high-bandwidth laser links in 2027. These tests will help determine whether space-based computing can move beyond an experiment and become a practical part of future AI infrastructure.