AI computing power "escaping Earth"? Google announces the "Project Sunroof," planning to deploy AI data centers in space by 2027

Google is attempting a groundbreaking infrastructure experiment, planning to relocate energy-intensive AI data centers to space to address power shortages and planning bottlenecks on the ground.

According to a report by the Financial Times on the 17th, Google has disclosed a plan called "Project Suncatcher," aimed at building a solar-powered space data center prototype. The project is not about establishing a single orbital monolith, but rather a cluster of 81 satellites equipped with AI chips that will fly in coordination in space and process data. As the first step of this plan, Google will collaborate with satellite company Planet, expecting to launch two prototype satellites into low Earth orbit by 2027.

The core logic of this move lies in leveraging the unique environmental advantages of space—particularly the nearly constant solar energy provided by sun-synchronous orbits, as well as the exemption from land and water resources required for construction on the ground. For investors, this releases a clear signal: despite facing limitations in the physical world, the tech industry is still trying to prove that AI has unlimited scalability. If AI models like Gemini can process queries in space and send results back to Earth, it will fundamentally change the cost structure and energy dependence of computing infrastructure.

However, this grand vision faces severe scrutiny regarding its technical feasibility and economic viability. Following Microsoft's previous termination of its underwater data center project (Project Natick), the industry remains cautious about the maintenance difficulties and cost-effectiveness of data centers in extreme environments. In addition to the high launch and maintenance costs, the risks of radiation, debris collisions, and increasingly crowded orbital traffic in the space environment cast a shadow over the prospects of this "ascension" plan.

Escaping Earth: Google's "Sun Chaser" Plan

According to a preprint paper released by Google researchers last month, Project Suncatcher envisions operating in a sun-synchronous orbit approximately 650 kilometers from Earth. This orbit can ensure a nearly uninterrupted power supply through solar panels mounted on the satellites, thereby powering energy-intensive AI workloads.

Unlike traditional ground data centers, this architecture relies on the collaborative work of a satellite swarm. These satellites not only need to handle computational tasks but also maintain communication while flying at high speeds. This solution aims to circumvent many obstacles faced by ground data centers, including increasingly complex planning approval processes, local community opposition, and the existing power grid's inability to meet the enormous energy gap created by the rapid expansion of AI. Industry estimates suggest that the energy demand generated by AI can no longer be met within the existing resource boundaries on Earth.

Space Minefield: Collision Risks and Technical Challenges in Crowded Orbits

Although space offers unlimited energy potential, its technical risks cannot be overlooked. Mojtaba Akhavan-Tafti, a space scientist at the University of Michigan, points out that the orbit planned for use by Google is one of the most crowded paths in low Earth orbit. To achieve interconnectivity, the distance between these satellites is only 100 to 200 meters Mojtaba Akhavan-Tafti warned that at such close distances, the margin for error in orbital navigation is almost zero. Once a single collision occurs, it could not only destroy a satellite but also trigger a chain reaction that leads to the destruction of an entire satellite constellation, scattering millions of fragments into an already "minefield" orbit. According to the European Space Agency, there are currently over 1.2 million pieces of debris larger than 1 centimeter in orbit, any one of which could cause catastrophic damage. With the surge in orbital traffic, the likelihood of the "Kessler effect"—where debris triggers chain collisions rendering orbits unusable—is increasing.

Additionally, Google's paper also acknowledges that space radiation degrades electronic devices and corrupts data. More challenging is that, unlike ground facilities, space data centers can hardly perform remote hardware maintenance, and how to handle faulty hardware or decommissioned satellites remains an unsolved problem.

Microsoft's underwater project falters amid high "space" costs

Before Google, Microsoft also explored unconventional environments to solve data center cooling and energy consumption issues. In 2018, Microsoft submerged an underwater data center named Project Natick off the coast of Scotland. However, reports indicate that the project has ended, and Microsoft stated that there are no further underwater plans.

Compared to underwater, establishing data centers in space presents exponentially greater challenges. Although the cost of rocket launches is continuously decreasing due to efforts from companies like SpaceX, at this stage, the unit cost of space electricity is roughly equivalent to that of ground electricity, with no significant cost advantage. When asked for specific comments, Google only cited relevant preprint papers and blog posts.

The space land grab by tech giants?

In addition to technological and cost barriers, the project has raised concerns about space governance. Astronomers worry that more satellite constellations will further interfere with scientific observations. Currently, there are nearly 16,000 satellites operating around the Earth, with nearly 9,000 belonging to Musk's Starlink network, and plans for another 15,000 new satellites are under review. Starlink has been criticized for interfering with optical images due to reflected light and for radio leaks disrupting radio astronomy.

The current space domain is gradually becoming an arena for tech giants like Jeff Bezos and Musk, lacking regulatory constraints. Google's plan not only risks exacerbating light pollution and radio interference but also exposes the absence of effective governance mechanisms for this public resource in space. For the market, this represents both a gamble showcasing the limitless potential of AI and a metaphor for the industry's "overexpansion" amid energy anxiety