Project Suncatcher: Google wants to put its AI data centers in solar orbit

Project Suncatcher: Google aims to place its AI data centers in solar orbit

Google has just unveiled one of its boldest projects to date: Project Suncatcher, a research initiative aimed at placing artificial intelligence data centers in orbit, powered by solar energy from space.

The goal? To find a sustainable solution to the immense energy needs of generative AI, which is driving electricity consumption on Earth to unprecedented levels.

Project Suncatcher: A Solar… Space Bet

According to Google, space could become “the best place to scale the computing power of AI,” explains Travis Beals, a senior director in the Paradigms of Intelligence division. In a blog post and a technical paper released this week, the company elaborates on its vision: “In orbit, satellites can capture almost continuous solar energy, eight times more productive than on Earth.”

Specifically, Google envisions deploying its Tensor Processing Units (TPUs)—the chips powering AI models like Gemini—on satellites equipped with high-efficiency solar panels, capable of providing electricity around the clock.

The idea is to harness a clean and nearly unlimited energy source, free from the spatial, cooling, or energy cost constraints of terrestrial data centers.

Orbital Data Centers

Project Suncatcher aspires to create actual constellations of interconnected satellite servers to perform large-scale AI computations. However, to compete with terrestrial data centers, several monumental challenges arise:

  1. Satellite Communication: Connections must support data rates of several dozen terabits per second.
    Google plans to maintain its satellites within less than a kilometer of each other to minimize latency—a novel and risky arrangement due to space debris.
  2. Radiation Resistance: The Trillium TPU chips have been tested in the lab and can withstand radiation levels equivalent to five years of space missions without major failures.
  3. Astronomical Costs (for now): Launching remains the primary hurdle. However, internal cost analysis suggests that operating an orbital data center could become “comparable” in cost per kilowatt per year to a terrestrial center by the mid-2030s.

A First Mission Scheduled for 2027

Google plans to collaborate with Planet, a company specializing in satellite imaging, to launch two satellite prototypes as early as 2027. These devices will carry miniaturized TPUs to test the feasibility of distributed computing in orbit and energy stability.

This test flight will serve to validate the hardware’s resilience, communication stability, and the energy efficiency of the concept.

Challenges: Between Innovation and Risks

Project Suncatcher emerges in a context where AI imposes increasing pressure on global energy infrastructures. Data centers already consume more electricity than some European countries, and this demand could triple by 2030.

However, moving AI to space comes with its own set of consequences:

  • Increased orbital collision risks due to the mass deployment of satellites,
  • Data sovereignty and security issues,
  • Dependence on the private space industry for maintenance and launches.

With Project Suncatcher, Google revives the spirit of its “moonshot” projects: ambitious, futuristic, and almost utopian. While the technical feasibility remains to be proven, the idea of relocating intensive computations into space opens a groundbreaking route for the tech industry.

What if, in ten years, our ChatGPT or Gemini queries were handled… 36,000 km above our heads?


Scroll to Top