Google is exploring a future in which some AI data centers are not built on land at all. Its Project Suncatcher concept aims to move AI compute into orbit, using solar-powered satellites connected by high-speed optical links.
The idea arrives as the tech industry races to expand AI infrastructure on Earth. Large data centers bring high energy costs, difficult logistics, and growing concern from communities that must live near facilities that are noisy, power-hungry, and demanding on water.
Why Google is looking beyond Earth
Project Suncatcher is Google’s answer to a basic constraint: AI compute needs enormous infrastructure, and that infrastructure is becoming harder to scale cleanly on the ground. The company has described space as potentially “the best place to scale AI compute.”
The timing matters because orbit is already becoming more useful for large technical systems. Satellite constellations like Starlink have shown that Internet communication can be relayed through orbital networks. Google is now studying whether a similar logic can be applied to high-performance AI accelerators.
The company’s proposed network would use satellites carrying TPUs, Google’s AI-focused processors. Those satellites would be powered by sunlight and linked together through free-space optical connections, forming a distributed data center above Earth rather than a conventional facility on the surface.
The solar advantage behind Project Suncatcher
A central part of the plan is energy. Google’s pre-print study describes placing the satellites in a dawn-dusk sun-synchronous low-earth orbit. That orbit would give them almost constant sunlight exposure, which explains the Suncatcher name.
The source article notes that even moving terrestrial data centers to solar power would not fully solve the problem. In orbit, Google says solar panels are up to eight times more efficient than they are on the surface of Earth.
That difference is important for AI compute because processing power depends on available energy. More consistent sunlight and higher panel efficiency could make orbital data centers attractive if the rest of the system can be made practical.
The potential benefits are straightforward:
- Access to almost constant sunlight in the planned orbit.
- Higher solar panel efficiency than on Earth.
- Less dependence on terrestrial land, power, and water resources.
- A possible path around community opposition to new data center construction.
The hard part is keeping satellites connected
Power is only one piece of the system. A data center also needs its many computing nodes to communicate at extremely high speeds. On Earth, that work is handled by fast optical interconnect chips inside facilities.
In orbit, Google would need wireless links capable of operating at tens of terabits per second. Early testing on Earth has already demonstrated bidirectional speeds up to 1.6 Tbps, and Google believes that can be scaled over time.
Distance creates another constraint. Received power falls with the square of distance, so the satellites would need to remain within a kilometer or less of one another. That would make the formation tighter than any currently operational constellation.
Google’s models suggest the idea may still be workable. Satellites several hundred meters apart would require only “modest station-keeping maneuvers,” according to the company’s analytical work described in the source.
Radiation is the test for ordinary hardware
Another challenge is hardware durability. Space equipment is often more expensive and less capable than terrestrial hardware because it must be hardened against radiation and extreme temperatures.
Google’s approach is different. Project Suncatcher is based on reusing components already used on Earth, even though those parts may be less robust inside a satellite. The source points to examples such as the Snapdragon-powered Mars Ingenuity helicopter as evidence that off-the-shelf hardware can sometimes survive longer than expected beyond Earth.
For Google, the crucial test is whether TPUs can operate long enough in orbit. The company says Suncatcher only works if TPUs can run for at least five years, equal to radiation exposure of 750 rad.
To study that risk, Google has been testing its latest v6e Cloud TPU, known as Trillium, with a 67MeV proton beam. The memory was the most vulnerable part, but the tests showed that TPUs could handle about three times the required radiation, almost 2 krad, before data corruption appeared.
What has to happen next
Google hopes to launch a pair of prototype satellites with TPUs by early 2027. Those first AI orbiters are expected to be expensive, so the early phase is less about immediate economics and more about proving the architecture.
The longer-term plan looks toward the mid-2030s. Google is planning for a period when launch costs are projected to fall to as little as $200 per kilogram. At that level, the source says space-based data centers could become as economical as terrestrial versions.
Project Suncatcher is still a moonshot. Google itself compares the timeline to its self-driving car work, which began 15 years ago and eventually led to Waymo vehicles that are almost fully autonomous today.
The same kind of long development path may apply here. The promise is large: solar-powered orbital compute that could ease pressure on land, power, and water. The obstacles are just as large: launch economics, formation flying, radiation tolerance, and networking speeds that must approach the demands of an AI data center.
If those pieces come together, Project Suncatcher could change where the next generation of AI infrastructure is built. Instead of only asking which community will host another data center, the industry may also ask whether some of that compute belongs in orbit.