Google Plans First Orbiting Data Center By 2027

Until now, the business of trying to put data centers in space has been on the speculative fringes of the digital infrastructure landscape — exclusively the territory of futurists and long-shot startups.

But the concept of space-based data centers has suddenly gained traction among tech's biggest names.

On Tuesday, Google announced Project Suncatcher, a program that aims to create an artificial intelligence data center in low Earth orbit, with computing spread across a tightly knit array of satellites.

While Google is funding the projects as a long-term “moonshot,” the company is devoting resources toward the effort. A pair of prototype satellites carrying Google’s proprietary AI chips are scheduled to launch by early 2027. 

Travis Beals, Google’s senior director for paradigms of intelligence, wrote in a blog post Tuesday that a solar panel in the right orbit can produce eight times more power in space than on Earth, producing continuous energy without a battery.

“In the future, space may be the best place to scale AI compute,” Beals wrote.

Google is hardly the first company to make concrete progress toward space-based data centers — two startups have launched data center prototypes into orbit and to the moon this year.

But these prior efforts have largely been treated skeptically as something of a novelty in the digital infrastructure ecosystem. Project Suncatcher marks an inflection point in what has been a sudden surge of interest in space-based AI data centers from the world's most valuable companies.

Earlier this month, Elon Musk wrote on X that SpaceX will eventually deploy space-based data centers. In October, Amazon and Blue Origin founder Jeff Bezos told a tech conference that he expects gigawatt-scale data centers in space in as little as 10 years. The idea was also championed by OpenAI CEO Sam Altman on a podcast earlier this year, while former Google CEO Eric Schmidt recently said his acquisition of Relativity Space was specifically motivated by a desire to put data centers into orbit.

Big Tech’s sudden interest, and now real investment, in space-based data centers is lending credibility to the nascent industry and its thesis that the ideal real estate for data centers is in orbit.

Rather than skyrocketing power consumption from terrestrial data centers causing power constraints and driving up energy prices, Beals and other proponents say AI training should be taken off the power grid and placed on solar-powered data centers in space.

While inconsistent energy production due to weather or time of day makes solar a bad fit for traditional data centers, orbiting solar panels are unhindered by weather, generate power more efficiently due to the lack of interference from the Earth’s atmosphere, and can be placed in an orbit where they are exposed to the sun at all times. 

“These giant training clusters, those will be better built in space because we have solar power there 24/7,” Bezos told an Italian tech conference last month. “There are no clouds and no rain. No weather.”

Beyond the appeal of free energy, the advent of generative AI created the first large-scale commercial use case for space-based computing. 

Space is ill-suited for most traditional, latency-sensitive workloads housed in terrestrial data centers. Quick data transfer is unreliable from orbit. Consequently, the handful of startups pursuing space data centers have largely targeted niche, latency-insensitive applications like disaster recovery storage or processing data generated by remote imaging satellites in space.

But now, training large language AI models has created an unprecedented wave of demand for workloads where latency isn’t a factor. It is a shift that has already driven data center development to remote locations like Wyoming and North Dakota that would have made little sense previously. Advocates for orbital data centers say space is the next logical step. 

Starcloud, formerly Lumen Orbit, is the first company to focus specifically on space-based AI training. The startup, which emerged from stealth last year, is backed by Y Combinator and has raised more than $12M to pursue its vision of a 5 GW orbiting data center.

This concept design is essentially a space station in low Earth orbit. Its computing equipment resides in modular pods that can be attached to or detached from the main craft, and the necessary power is generated by a vast array of solar panels stretching 4 kilometers.

The firm launched its first test satellite earlier this month: a dishwasher-sized unit containing Nvidia graphics processing units. 

Google’s Project Suncatcher is taking a fundamentally different design approach. Instead of building a single large data center, Google intends to distribute AI computing across a cluster of 81 satellites spread over a square kilometer. Musk has suggested SpaceX may take a similar approach by modifying its Starlink satellites.

While creating reliable high-bandwidth connectivity between the satellites could be an obstacle, the project’s leaders say these distributed data centers will have significantly lower launch and construction costs and will be easier to scale up compared to the single-structure designs being pursued by Starcloud. 

While it may be a moon shot, they argue it is the most realistic path toward making large-scale AI data centers in orbit a reality. 

“By focusing on a modular design of smaller, interconnected satellites, we are laying the groundwork for a highly scalable, future space-based AI infrastructure,” Beals wrote.