The search giant is shipping its artificial intelligence hardware into orbit, and it has now spelled out precisely how it intends to make the plan work. Project Suncatcher, a grand scheme unveiled last year, calls for launching a prototype satellite carrying the company’s own internal AI work aboard a SpaceX vehicle’s Transporter-18 rideshare mission. The payload carries Tensor Processing Units, or TPUs, the custom chips that power much of Google’.
Google has set its sights on building orbital data centers capable of running massive AI workloads without pause. Rather than depending on the Earth’s rotation for power, these facilities would draw nearly constant energy from the sun instead. The company sees space-based supercomputing as a practical prospect within a decade, a wager on whether such systems can be made to work in orbit.
Hardware Survival
The journey of a rocket through low Earth orbit takes roughly 10 minutes. Throughout that time, the vehicle endures heavy shaking and continuous pressure from acceleration, reaching up to 10 times the force of gravity, or g-force. Some parts, including the TPU chips, face even higher pressures, ranging from 50 to 100 g.
To simulate the forces of a rocket launch, the team subjected the satellite to intense vibration testing, shaking it along all three axes. Such tests seldom proceed without incident, yet the hardware endured the force applied to it. This outcome is a pleasant surprise, not a guarantee of future performance.
After the TPUs get into space, the issue becomes radiation from solar events and cosmic rays, which can damage electronics. At UC Davis’s Crocker Nuclear Laboratory, Google put TPUs through a proton beam test while running AI workloads, monitoring closely to see how errors, such as a bitflip, would impact those workloads.
The Trillium TPUs have proven surprisingly durable so far, able to withstand a radiation total ionizing dose exceeding what they would experience on a five-year space journey. Still, certain tests can only be run in orbit, and the pending launch will finally supply that missing information.
Cooling in Space
A great deal of heat builds up within a narrow space inside TPUs, a danger that calls for safe diffusion before the chips overheat. In the absence of air, however, space offers no natural current to carry that heat away. Within a vacuum, the only means of heat diffusion is through radiators, a distinct departure from the usual method of cooling electronic components.
The company is building a cooling setup for its chips that joins heat pipes with radiators. Its team put the technology through testing inside a thermal vacuum chamber that recreates the temperature and vacuum conditions found in space. Google will observe how the new TPU cooling system behaves in orbit and make design improvements as it gathers more knowledge.
Satellite Interconnectivity
The next generation of satellites will each carry dozens of TPU chips while circling the Earth in groups. Each satellite must know its own position and how it sits relative to its neighbors in order to keep the bandwidth needed for processing AI. Lasers will be used for communication between satellites to achieve this.
Space technology is already here, though most cutting-edge systems are built for low bandwidth across vast distances. Google’s lasers must work at very high bandwidth over extremely short distances instead. Keeping that connection steady demands extraordinary precision, much like hitting a coin-size target from miles away with both points in motion.
Google will test this work in 2027 when it puts two satellites in orbit.
What Happens Next
The initial flight serves as a test to spot what functions and where things go wrong, so the lessons can be used on later journeys. Google is not declaring triumph now; it is gathering information instead.
The company’s video series examines the science driving the mission, with the team addressing what they hope to discover and the engineering obstacles still to come. The upcoming goal is 2027, when the company will launch two satellites to put the laser interconnectivity to the test.
“Exploring space as a viable location for scalable AI compute won’t happen all at once.”
The Project Suncatcher team offers a candid self-assessment, acknowledging that its approach rests on careful engineering work, beginning with demonstrating that hardware can endure the physical and unpredictable realities of orbital operations.
Key Facts Box
- Launch: Prototype satellite on SpaceX’s Transporter-18 rideshare mission
- Hardware: Tensor Processing Units (TPUs)
- Radiation testing: Proton beam facility at UC Davis’s Crocker Nuclear Laboratory
- Cooling testing: Thermal vacuum chamber
- Next milestone: Two satellites in orbit, 2027
- Radiation tolerance: Survived a dose greater than five-year space mission exposure
Why This Matters
Among a small number of firms with the means to construct, evaluate, and release the hardware, Google stands as one of them. The consequences are significant. Should AI infrastructure prove capable of running reliably in space, it alters the financial basis of cloud computing.
This broader push is framed by the company as a way to make sure AI’s huge benefits in fields such as healthcare and scientific discovery reach everyone far into the future. That is the stated aim. The route to it takes in vibration chambers, proton beams, and two satellites in orbit.
A real attempt at something new is underway here. Rather than merely discussing space-based AI, Google has placed the hardware itself into orbit to test whether it functions.
Source material: “Google’s Project Suncatcher to put ML infrastructure in space,” Google.
Get the Notebook.
The day's best stories and every fresh verdict, in plain English, in your inbox by seven. One email a day, no more.