Google’s Project Suncatcher prototype reached orbit, made contact and was operating as expected, Google said on October 1, 2026. The company has begun testing how its AI processors handle spaceflight stress, radiation and extreme temperatures—a research step, not an orbital data center serving customers. Google’s announcement describes the project as long-term research. NeoTeo previously covered Google’s initial contact with the Project Suncatcher satellite; the bigger question is what it would take to turn a hardware test into useful infrastructure.
What Project Suncatcher is testing
Google built the prototype with Planet and launched it on SpaceX’s Transporter-18 rideshare mission. Its experiment is focused on Tensor Processing Units (TPUs), Google’s processors for AI workloads, and how they respond to the physical stresses of spaceflight, radiation and thermal extremes.
That is a meaningful step: hardware is in orbit and the research can examine conditions that are difficult to reproduce fully on the ground. But a prototype test is not a network of computing facilities. Business Insider España’s interviews with José Fernández Tamames and Paula Raboso describe current orbital-computing activity as demonstrations and small experiments, rather than commercial-scale infrastructure.
What stands between a test and a useful facility
An orbital data center would need to power its computers, remove their waste heat, keep hardware reliable and move data to where it is needed. Solar arrays could supply power, but cooling still poses a design challenge: in a vacuum, heat cannot escape through surrounding air, so radiators must release it as radiation. Those radiators add hardware and mass that must be launched.
Business Insider España’s interviews with Fernández Tamames and Raboso also raise radiation-related reliability, launch costs, repair and replacement access, environmental effects and security. A terrestrial facility can be reached by technicians; equipment in orbit presents a different maintenance problem. Debris adds another concern for a system that depends on many satellites.
What the five-year cost model says
An IEEE Spectrum scenario compares two hypothetical systems with 1 gigawatt of capacity. Its orbital design uses roughly 4,300 satellites and is estimated to cost about $51.1 billion over five years, including design, construction, launch and operating expenses. The comparable terrestrial scenario is estimated at about $16 billion for the same period. IEEE Spectrum’s model puts the orbital case at roughly 3.2 times the terrestrial cost, calculated from those rounded estimates.
| Scenario | Capacity used for comparison | Modeled cost over five years |
| Orbital network, roughly 4,300 satellites | 1 GW | About $51.1 billion |
| Terrestrial data center | 1 GW | About $16 billion |
These are scenario estimates, not a construction bid or an observed operating cost. They give a defined comparison for one modeled scale; they do not establish what a future orbital facility would cost. A lower launch price in the future would affect one part of the equation, while heat management, hardware replacement and operations would remain part of the broader infrastructure challenge.
Which computing jobs might fit in orbit?
The best early use may depend on how much data a job needs to exchange with Earth. In an October 24, 2025 analysis, Andrew Cote argues that processing Earth-observation data in space, along with some less-interactive inference or isolated training workloads, could fit communication constraints better than cloud services requiring continuous, high-volume exchanges with users on Earth. That is an engineering assessment, not a demonstrated service or proof of commercial demand.
For orbital computing, the key question is therefore not just whether a processor can run in space. It is whether a useful workload can be powered, cooled, maintained and connected at a cost that makes the whole system practical.
