Google has begun testing its Trillium AI chips in low Earth orbit, opening a feasibility path toward satellite-based data centers. The hardware launched on October 1st, 2026, when a SpaceX Falcon 9 carried the Project Suncatcher prototype satellite into orbit with four Trillium TPUs on board.

The immediate goal is narrow: determine whether the chips can survive launch vibrations, radiation, and extreme thermal conditions, and still operate in space. Google and satellite developer Planet plan a second phase in 2027, launching two purpose-built satellites to test high-bandwidth optical communication between spacecraft.

The experiments target AI's power problem. Terrestrial data centers consume large amounts of electricity for both computation and cooling. Google's researchers estimate that appropriately positioned solar panels in orbit could generate up to 8 times more power than equivalent panels on Earth, while placement in space removes the need for conventional air-based cooling entirely.

Scaling demands 370,000 tons of payload

The architecture Google envisions consists of clusters of satellites carrying large numbers of TPUs. One proposed configuration places 81 satellites in a cluster within roughly a 1-kilometer radius.

Reaching useful scale would require sending 370,000 tons of payload into orbit, according to TechCrunch. If SpaceX's Starship eventually delivers 200 metric tons per launch, that translates to 1,850 launches over 10 years, or 185 launches per year. Starship has never flown more than 5 times in a single year to date.

Launch cost presents an equally hard constraint. Sending a kilogram of hardware to low Earth orbit currently costs around $3,600. Google calculates the price must fall to $200 per kilogram for orbital computing to become roughly comparable with terrestrial data-center power costs. The company estimates that price point could arrive by the mid-2030s.

Competitors are moving in parallel

Other projects may accelerate the cost decline by building launch demand and competing on architecture. Blue Origin has filed plans for up to 51,600 orbital data-center satellites under Project Sunrise. Starcloud is working toward launching up to 88,000 satellites and has already demonstrated an Nvidia H100 GPU operating in orbit. Axiom Space is pursuing its own orbital data centers and reports its technology is scaling from kilowatts toward megawatts.

The technical and economic challenges remain large. Radiation hardening, thermal management, inter-satellite bandwidth, and debris risk all sit outside the scope of the current prototype mission. The first Suncatcher flight tests component survival, not economics.

Google frames the effort accordingly. The company calls Project Suncatcher a "long-term research moonshot." The 2027 optical-communication satellites will provide the next data point on whether clustered orbital compute can move from concept to engineering program. Proof that the launch economics work — at $200 per kilogram and hundreds of Starship flights per year — remains years away.