Google Announces Ambitious Plan To Launch AI Datacenters Into Space

Key Takeaways

Google has launched an ambitious initiative to build artificial intelligence datacenters in orbit, announcing Tuesday that it plans to send solar-powered satellites carrying its proprietary AI processors into space as early as 2027.

The project, dubbed Project Suncatcher, represents Google's response to the mounting energy demands of AI development, which has sparked growing concerns about the environmental impact of massive terrestrial datacenters.

Tech companies are projected to spend $3 trillion on Earth-based datacenters globally, raising alarms about carbon emissions and resource consumption.

Travis Beals, Senior Director for Paradigms of Intelligence at Google, emphasized the potential of space-based computing in a blog post announcing the initiative.

"In the future, space may be the best place to scale AI compute," Beals wrote. "Working backward from there, our new research moonshot, Project Suncatcher, envisions compact constellations of solar-powered satellites, carrying Google TPUs and connected by free-space optical links."

Harnessing unlimited solar power in orbit

The core advantage of space-based datacenters lies in their access to nearly continuous solar energy.

Google plans to deploy its Tensor Processing Units on satellites operating in a dawn-dusk sun-synchronous low-Earth orbit approximately 400 miles above Earth's surface.

In this orbital configuration, solar panels can be up to eight times more productive than their terrestrial counterparts and generate power almost continuously, eliminating the need for heavy battery systems.

"If things keep going down the path where we keep having more uses for AI and we keep wanting more energy to power it, this has tremendous potential to scale," Beals told Semafor. "But as with all the moonshots, there's no certainty."

Google envisions constellations of approximately 80 satellites flying in tight formation—positioned within hundreds of meters of each other—connected via free-space optical links, transmitting data at tens of terabits per second.

This configuration represents a significantly denser arrangement than current satellite constellations like Starlink, which maintain a spacing of roughly 120 kilometers between satellites.

Technical hurdles and radiation testing

The ambitious project faces substantial engineering challenges. Space-based AI chips must withstand significantly higher radiation levels than their Earth-based counterparts, potentially causing hardware failures and data corruption through particle collisions with transistors.

To address these concerns, Google has conducted extensive testing on its Trillium-generation TPUs. The company used a particle accelerator at the University of California, Davis, to simulate years of space radiation exposure.

"They held up quite well," Beals said, indicating the processors could handle a five- or six-year mission. The tests showed that the chips "survive a total ionizing dose equivalent to a 5-year mission life without permanent failures."

Communication presents another critical challenge. Google achieved 800 gigabits per second, one-way transmission in laboratory demonstrations using a single optical transceiver pair.

However, maintaining these high-speed connections between satellites in close formation requires sophisticated station-keeping maneuvers to counteract gravity anomalies and atmospheric drag.

Economic viability on the horizon

While launching hardware into space remains expensive today, Google's cost analysis suggests the economics may soon shift in favor of orbital infrastructure.

The company projects that by the mid-2030s, assuming launch costs fall to $200 per kilogram, the energy costs of operating a space-based datacenter could be "roughly comparable" to an equivalent Earth facility on a per-kilowatt-year basis. Current launch prices exceed this target by more than tenfold.

"Our initial analysis shows that the core concepts of space-based ML compute are not precluded by fundamental physics or insurmountable economic barriers," Beals stated.

Google CEO Sundar Pichai acknowledged the significant work ahead. "Like any moonshot, it's going to require us to solve a lot of complex engineering challenges," Pichai said. "More testing and breakthroughs will be needed as we count down to launch two prototype satellites with Planet by early 2027, our next milestone of many."

Growing competition in orbital computing

Google is not alone in pursuing space-based AI infrastructure. The company faces competition from multiple players racing to establish orbital computing capabilities.

Startup Starcloud successfully launched the first datacenter-class GPU into space on November 3, 2025.

The 60-kilogram Starcloud-1 satellite, carrying an Nvidia H100 GPU, separated from a SpaceX Falcon 9 rocket and is now operational in orbit.

Philip Johnston, Starcloud's co-founder and CEO, celebrated the milestone on LinkedIn: "Starcloud was founded only 21 months ago, and this satellite hosts the first Nvidia H100 in space, which will provide high-powered inference and fine-tuning capabilities for other satellites."

"In space, you get almost unlimited, low-cost renewable energy," Johnston told Nvidia's blog. "The only cost on the environment will be on the launch, then there will be 10x carbon-dioxide savings over the life of the data center compared with powering the data center terrestrially on Earth."

Johnston made an even bolder prediction: "In 10 years, nearly all new data centers will be built in outer space."

Other tech leaders have also expressed interest in orbital computing. Jeff Bezos, Amazon's founder and owner of Blue Origin, predicted last month that gigawatt-scale datacenters will operate in space within 10 to 20 years. Former Google CEO Eric Schmidt reportedly acquired rocket company Relativity Space with plans to deploy data centers in orbit.

Even Elon Musk, who controls SpaceX and Starlink, weighed in on Google's announcement. "Great idea lol," Musk commented on X. "Only possible because of SpaceX's massive advances in launch technology!" Pichai responded to Musk's comment, acknowledging the role of improved launch capabilities in making such projects feasible.

First test flight scheduled for 2027

Google plans to partner with satellite imaging company Planet Labs to launch two prototype satellites by early 2027.

Each prototype will carry four TPUs and will serve as a crucial test of whether distributed machine learning workloads can run effectively across satellite constellations in the harsh conditions of low-Earth orbit.

The mission will validate Google's models for satellite formation flying, test the durability of TPUs in the space environment, and demonstrate the viability of free-space optical links for high-bandwidth communication between satellites.

If successful, Project Suncatcher could fundamentally reshape where and how AI computation happens, offering a path to scale artificial intelligence development while reducing the strain on Earth's energy and water resources.

Read more: