Google CEO Sundar Pichai says data centers in space are coming

Google spent ten years turning the moonshot into a brand. Now the company is drawing a line to the real Moon.

Alphabet CEO Sundar Pichai now openly supports data centers in space, Fox News reports in a Sunday interview that the company will begin producing AI hardware in 2027 and that in about 10 years, extraterrestrial infrastructure will become a common way to fuel the AI ​​boom.

“I have no doubt that in a decade or so we will see this as a more normal way of building data centers,” he said. So this time, Pichai's moon rhetoric comes with a rocket attached. And in the not-too-distant future, Google's cloud could exist far beyond the atmosphere and beyond the planet forever.

“At Google, we always pride ourselves on being successful,” Pichai said. “One of our ideas is: How can we one day create data centers in space so that we can better harness the energy from the sun, which is a hundred trillion times more energy than we produce today on the entire Earth?”

Internally, the project exists under the very loud name Project Suncatcher. IN November blog postGoogle billed it as a “research project” to bring artificial intelligence computing into orbit, with swarms of solar-powered satellites using TPUs and communicating with each other via laser links rather than fiber optics. Essentially, the company hopes to “one day scale machine learning into space.”

The opening act is small—two test satellites in partnership with Planet by early 2027—but the sketches jump right into dense satellite clusters the size of city blocks, an early sketch of what tomorrow's overseas data centers might look like in the 2030s.

“We're taking the first step in '27,” Pichai told Fox News. “We'll send tiny racks of machines, put them on satellites, test them, and then start scaling from there.”

The cosmic presentation appears when the Earth begins to look like bad long term landlord For AI assembly. A Lawrence Berkeley National Laboratory 2024 Report. found that US data centers already consume about 4.4% of the country's electricity, a share that could rise to 12% by 2028 as the number of graphics farms increases. McKinsey sets a price tag on the race to scale data centers: By 2030, global capital spending on data centers will be about $6.7 trillion, of which about $5 trillion will be for AI-ready infrastructure. At some point, “just build another region” stops being a strategy and becomes an electrical engineering problem.

Google already lives inside this mathematics. The company's disclosures and external data show that electricity consumption at its data centers has more than doubled in four years, even as it tries to keep emissions low through carbon-free energy purchases. At the same time the company just committed $40 billion through 2027 to campuses in Armstrong and Haskell counties in Texas.

These are “regular” data centers. Suncatcher is a contingency plan for situations where even friendly networks and expedited permitting fail to meet the challenge.

The company's position is that space solves many of the headaches that its Texas builds create. Above the atmosphere, satellites can be exposed to near-constant sunlight, harvest much more solar energy from each panel, and dump waste heat into deep space instead of warm rivers and angry taxpayers. Launch costs continue to drop, and as part of Project Suncatcher, Google's research team has already envisioned constellations of tightly clustered satellites linked by laser links and filled with TPUs—a TPU orbital unit that looks more and more like a data hall with a better view. Pichai's “within a decade or so” line is essentially a bet that the rocket and chip cost curves will intersect quickly enough to make this architecture competitive with another concrete box in West Texas.

Pichai is not alone in thinking that the next phase of the cloud may exist beyond the planet. Tech executives became obsessed with the idea. Salesforce CEO Marc Benioff called space “The cheapest place for data centers” thanks to continuous solar power and no batteries is a reference to a video clip of Tesla CEO Elon Musk discussing orbital artificial intelligence and space-based data centers. Musk cites numbers about 300-gigawatt Starships, Amazon founder Jeff Bezos talks about a timeline of building data centers in space within 10 to 20 years, and OpenAI CEO Sam Altman speculates about “maybe” putting data centers in space—”maybe we'll build a big Dyson sphere,” he said.

Meanwhile, Y Combinator- and Nvidia-backed Starcloud startup was busy sending its own satellite equipped with artificial intelligence. into orbit and promise that extraterrestrial data centers could cut emissions by a factor of 10 compared to their terrestrial counterparts, even when accounting for rocket fuel. In this version of the space race, Mars is just the scenery. The real prize is the cheapest, cleanest marginal megawatt that can be specified in the language model.

There is a risk that Google's orbital extension will become a hugely expensive science fair project with gorgeous launch photos. But if the AI ​​boom requires power plants the size of Texas anyway, Google would prefer to keep some of them afloat. Most companies are worried about running out of runway. Google is playing out what will happen when it runs out of planets.