Google is on the verge of advancing its efforts to establish AI data centers in outer space.
On Thursday, the tech powerhouse revealed that its Project Suncatcher is set to launch a prototype satellite into low Earth orbit next week. This mission aims to evaluate the performance of Google Tensor Processing Units (TPUs) under the extreme conditions of space.
“This inaugural launch focuses on identifying operational strengths, pinpointing potential failure points, and utilizing these insights for future missions,” the company stated in a blog post.
Project Suncatcher represents Google’s ambitious long-term research initiative, first introduced last year, aimed at determining whether large-scale AI computing infrastructure can effectively function in space.
The fundamental concept involves equipping satellites with AI chips that could harness near-constant sunlight for power. These satellites would communicate with one another using lasers to manage substantial AI computational tasks. Google asserts that satellites in low Earth orbit could produce up to eight times more solar energy than what is achievable on Earth.
This innovation could alleviate some of the energy and environmental challenges confronting ground-based data centers, which are facing increasing scrutiny and criticism.
Google is not alone in its exploration of space as a potential solution. Companies such as SpaceX and Starcloud are also investigating the viability of orbital data centers. In November 2025, Starcloud successfully launched an Nvidia H100 GPU into orbit. Meanwhile, SpaceX has prominently featured space-based AI infrastructure in its IPO pitch earlier this year.
For its part, Google’s prototype satellite will be transported aboard SpaceX’s Transporter-18 rideshare mission next week. Developed in collaboration with aerospace firm Planet Labs, this mission will mark the first time four Google TPUs are sent into orbit.
The mission aims to gather critical data on the performance of the chips during the journey to space and their operation in low Earth orbit.
For instance, Google reports that the approximately 10-minute rocket ascent into low Earth orbit subjects the spacecraft to intense vibrations and forces that can be up to ten times greater than Earth’s gravity. Individual components, including the TPU chips, may briefly encounter forces ranging from 50 to 100 times stronger than normal gravitational pull.
Once in orbit, the chips will also contend with elevated radiation levels due to solar events and cosmic rays, which can compromise electronics or result in data errors known as bit flips.
Previously, Google conducted tests on its Trillium TPUs at UC Davis’s Crocker Nuclear Laboratory, exposing them to a proton beam while executing AI workloads. The team observed how errors, such as bit flips, impacted those workloads. According to Google, the chips performed “exceptionally well,” enduring a total radiation exposure surpassing what they would typically face during a five-year space mission.
AI chips generate significant heat, yet the vacuum of space lacks airflow. This absence means that conventional cooling systems used in terrestrial data centers cannot be employed.
“We are exploring various solutions for this challenge, including a combination of heat pipes and radiators to manage the cooling of the chips,” the company explained in its blog post. “So far, our team has trialed this technology in a thermal vacuum chamber that replicates the thermal and vacuum conditions found in space. We aim to evaluate the performance of our new TPU cooling system in space and refine our designs based on our findings.”
The insights gained from these experiments will significantly influence the design and objectives of future Project Suncatcher missions. Google has already announced plans to deploy two satellites into orbit by 2027, where they will test high-bandwidth lasers intended to connect clusters of satellites.
For the original content, including associated photos and images, please visit this source. We do not claim authorship of these materials; they are used strictly for informational purposes with appropriate attribution to the original source.









