Mirror Review
November 06, 2025
Google has announced Project Suncatcher, a research moonshot that explores whether the company can one day run large-scale machine-learning compute in space.
The idea is bold but straightforward: place Google’s TPU AI chips on solar-powered satellites, fly them in tight formations, connect them with extremely high-speed laser links, and eventually build something that behaves like a data center in orbit.
The first major step toward this vision is Google’s partnership with Planet to launch two prototype satellites by early 2027.
These early missions will test how the hardware behaves in space, how radiation affects the chips, and whether communication links can reach the speeds needed for AI workloads.
Although it’s still early research, Project Suncatcher signals a major shift in how AI compute could work in the future if energy, bandwidth, and orbital engineering align.
10 Amazing Google’s Project Suncatcher Facts
1. Project Suncatcher is Google’s plan to explore AI data centers in space
Google calls it a “moonshot” because the project takes on a tough problem with long-term potential.
The goal isn’t a near-term product. Instead, the team wants to understand what it would take to build AI compute clusters beyond Earth, powered mainly by the Sun.
In simple terms, Google wants to see whether space can host the next generation of data centers, where solar power is abundant and uninterrupted.
2. Two test satellites will launch by early 2027
To move from theory to practice, Google is working with Planet to launch two small satellites by 2027. These satellites won’t form a full data center, but they will help answer key questions:
- How do Google’s chips behave in space?
- Does the cooling system work?
- Can satellites communicate fast enough for AI tasks?
The Project Suncatcher mission is essentially Google’s first “hardware experiment” in orbit.
3. Satellites will sit in an orbit with almost constant sunlight
The plan is to place these satellites in a dawn–dusk sun-synchronous low Earth orbit. Here’s the simple explanation:
- They follow a path where they stay close to the sunlight line on Earth.
- That means continuous solar power.
- Continuous power means the satellites could support steady AI workloads without large batteries.
This orbit is key because solar panels in space can produce up to 8 times more energy than panels on Earth.
4. These satellites will fly unusually close to each other
Typical satellite constellations stay far apart, often tens or hundreds of kilometers.
But Suncatcher satellites would fly only 100–200 meters apart, closer than two football fields.
This unusual formation allows:
- Stronger laser signals
- Higher bandwidth
- Lower communication delays
Google’s modeling shows that, even at these close distances, satellites can remain stable with only minor adjustments.
5. Google already achieved 1.6 Tbps in lab tests — a huge jump in satellite networking
For space-based AI compute to work, satellites must communicate at speeds comparable to those in Earth data centers.
Google tested its optical link system in the lab using just one laser link and achieved 1.6 terabits per second.
That’s fast enough to transfer:
- an entire HD movie in milliseconds
- or huge AI model parameters instantly
This is a major reason Project Suncatcher is more than just an idea. Google has proven that some of the tech already works.
6. Google’s TPUs handled radiation better than expected
Radiation is one of the biggest challenges in space. It can corrupt memory or damage chips permanently.
Google tested its Trillium TPU using a 67 MeV proton beam to simulate space radiation. The results:
- The HBM memory showed issues only after 2 krad(Si)—almost three times more than the expected five-year dose.
- The TPU itself survived up to 15 krad(Si) with no hard failures.
This means Google’s commercial chips are far more radiation-tolerant than many expected.
7. Project Suncatcher combines three complex engineering fields
Suncatcher sits at the intersection of:
- Astronautics (orbital mechanics, satellite control)
- Optical communications (DWDM lasers, high-bandwidth links)
- AI systems (distributed training and inference)
Google’s research team is modeling everything from gravity perturbations to thermal behavior. It’s a rare project where space engineering and AI infrastructure meet at full scale.
8. Falling launch prices could make space compute affordable
One reason Suncatcher is even being considered: launch costs are dropping fast.
Google’s analysis shows that if launch prices fall below $200 per kilogram, running AI compute in space might be as cost-effective as running a traditional data center, especially when you factor in the free solar power in orbit.
That cost threshold could realistically be reached in the mid-2030s if industry trends continue.
9. Thermal management and ground links are still big unknowns
Google openly says that several problems remain unsolved.
- Cooling: Space is cold, but heat has nowhere to escape without air. High-power chips get hot fast.
- Ground connectivity: Even if satellites talk quickly to each other, they still need high-speed links to Earth.
- Reliability: Thousands of compute satellites must survive for years.
These are the engineering challenges that future prototypes will need to address.
10. This project could change the meaning of “data center”
If Suncatcher succeeds, it could redesign the idea of an AI data center.
Instead of massive buildings on Earth consuming gigawatts of electricity, future ML workloads might run in orbit, drawing almost unlimited solar power with fewer land and resource constraints.
Google compares the potential transformation to how system-on-chip (SoC) technology changed smartphones, integrating many components into one compact, efficient package.
That same type of integration could happen with compute + solar + cooling in space.
Conclusion
Project Suncatcher is still early, but its direction is clear.
“Artificial intelligence is a foundational technology that could reshape our world,” Google writes, adding that Project Suncatcher explores “where we can go next to unlock its fullest potential.” Those words underline that Suncatcher is as much a research program as a technical roadmap.
Google is testing whether we can move some of the world’s heaviest AI workloads off the planet and into sunlight-rich orbits.
With TPUs that survive radiation, optical links that already hit 1.6 Tbps, and prototype satellites planned for 2027, the project is grounded in real research and engineering.
If Google solves the remaining challenges, Suncatcher could reshape how future AI data centers are built and where they are built.
