Google announced this week that it will launch its first prototype satellites to test Tensor Processing Units in space, marking a major step toward orbital AI infrastructure. The project, called Project Suncatcher, represents Google’s bet that space-based computing could solve the growing energy demands of artificial intelligence at scale.
Google wrote:
Exploring space as a viable location for scalable AI compute won’t happen all at once. It takes methodical engineering, starting with proving our hardware can handle the physical and unpredictable realities of operating in orbit. This first launch is about seeing what works, identifying points of failure, and applying those findings to future missions.
The initial test satellites launch within days through SpaceX’s Transporter-18 rideshare program, with Google partnering with satellite company Planet Labs for deployment. Full prototype testing moves forward in early 2027. The initiative aims to demonstrate whether constellations of solar-powered satellites equipped with Google’s specialized AI chips can operate sustainably in orbit.
The energy calculus driving this moonshot is compelling. Solar panels at orbital altitudes receive up to eight times more energy annually than Earth-based systems. The sun itself emits more power than 100 trillion times humanity’s current electricity production. Space offers near-perpetual sunlight for satellites positioned in dawn-dusk orbit, eliminating battery dependency and enabling continuous power generation without connection to terrestrial grids.
Google’s approach centers on its Trillium v6e Cloud TPU, a custom-built machine learning accelerator chip. Researchers tested these chips in a 67 megaelectronvolt proton beam to assess radiation tolerance. Results show the TPU can withstand radiation exposure equivalent to five years of orbital operation without permanent damage, critical validation for multi-year mission duration.
However, multiple engineering barriers remain. Thermal management presents immediate challenges. Traditional air cooling becomes impossible above the atmosphere. Google is testing heat pipe and radiator combinations to dissipate computation-generated heat into space. Early bench tests confirmed 1.6 terabits-per-second bidirectional transmission using single optical transceiver pairs, validating laser-based communication between satellites.
As Google states:
Once the TPU chips make it to space, the level of radiation outside the Earth’s atmosphere presents another challenge to overcome. Solar events and cosmic rays can wreak havoc on electronics, so our team tested TPUs in a proton beam facility at UC Davis’s Crocker Nuclear Laboratory while running AI workloads. During the test, we monitored closely to see how errors, like a bitflip, would affect our workloads. Initial results have shown that our Trillium TPUs hold up remarkably well, and can survive a radiation total ionizing dose greater than what they would receive during a five-year space mission.
We’re working on a number of different approaches for this, including a combination of heat pipes and radiators to cool the chips. So far, our team has tested the technology in a thermal vacuum chamber that simulates both the thermal and vacuum environment in space. We’ll see how our new TPU cooling system works in space and refine our designs as we learn more.
Orbital dynamics add complexity. Simulations using Hill-Clohessy-Wiltshire equations and JAX-based differentiable modeling suggest that an 81-satellite cluster at approximately 650 kilometers altitude would require only modest station-keeping to maintain 100-200 meter spacing. Launches at this altitude orbit the Earth in roughly 90 minutes, balancing communication latency against atmospheric drag.
The competitive landscape shows Google is not alone in exploring space infrastructure. SpaceX and Starcloud have announced parallel initiatives. Meanwhile, China recently opened the world’s first underwater data center in Hainan Province, reflecting broader urgency around finding alternative locations for AI compute.
Google’s timeline projects cost parity between orbital and Earth-based data centers by the mid-2030s. That projection depends on solving launch economics, currently the dominant constraint. Rideshare programs like SpaceX’s Transporter flights reduce marginal costs, but building satellite constellations at scale remains capital-intensive.
Earlier validation at UC Berkeley confirmed TPU functionality and thermal behavior in controlled environments. Orbital testing now becomes essential because Earth-based labs cannot replicate space radiation, vacuum conditions, or true launch dynamics. Failed silicon in orbit costs far more than failed chips on ground, amplifying the stakes for every component qualification.
Google’s researchers emphasized this remains exploratory. Realizing orbital AI infrastructure requires solutions to power distribution architecture, inter-satellite networking at latency-critical performance levels, and operational resilience across thousands of hours of unattended orbital operation. The company framed Project Suncatcher as a decade-long research effort, not a near-term commercial service.
Yet the underlying logic is practical. AI model training and inference consume exponentially growing electricity. Hyperscalers compete for land suitable for data center construction. Water availability constrains cooling. However, space offers freedom from those constraints. That is, if engineering challenges can be overcome.
