As the demand for AI (artificial intelligence) computing power increases exponentially, and ground data centers face bottlenecks such as energy and heat dissipation that are difficult to break through, Silicon Valley technology giants have recently set their sights on space. Building data centers in space has become the next battlefield in their AI competition.

Google CEO Sundar Pichai talked about the hot topic of "building data centers in space" on a podcast this week.

"Obviously, this is an ambitious and bold plan," Pichai said. He admitted that the idea seems "crazy" now, but "when you really step back and think about it and imagine the scale of computing power we will need in the future, it starts to make sense. It's just a matter of time."

Pichai's statement refers to "Project Suncatcher," a new long-term research plan announced by Google recently. The goal of Suncatcher is to “one day deploy machine learning at scale in space,” according to a company blog post.

Pichai did not reveal too many details, but said, "Hopefully, in 2027, we can deploy TPU (Tensor Processing Unit) somewhere in space." TPU is Google's custom AI chip.

“Maybe we’ll get a Tesla Roadster,” he joked.

The joke in Pichai's words refers to a "feat" performed by Elon Musk that year: Musk fixed his old Tesla Roadster sports car on a SpaceX rocket and launched it into orbit. There was also a mannequin in a space suit on the driver's seat. The sports car was launched in 2018 and was still floating in deep space as of earlier this year, when astronomers mistook it for an asteroid.

But in the AI ​​era, the space ambitions of Musk and other technology giants are no longer comparable to the gimmick of this Roadster sports car.

"Starship should be able to deliver about 300 gigawatts of solar-powered AI satellites to orbit per year, and maybe even 500 gigawatts. The 'per year' magnitude is the key to this," Musk wrote on X earlier this month.

The numbers Musk is talking about represent unprecedented data center capacity. Goldman Sachs said earlier this year that global ground data center power capacity currently stands at 59 gigawatts.

Global electricity demand is expected to double by 2050, driven in part by a boom in building AI data centers. In the United States, data centers are the biggest driver of the surge in electricity demand, putting huge pressure on the country's power grid.

Musk, Pichai and other leaders in the technology field have realized that the power demand brought by AI data centers will be difficult to meet. Because of this, they are eager to build data centers into space.

Amazon founder Jeff Bezos predicts that data centers will be in space within the next 10 to 20 years.

“I do think that over time the world will be filled with data centers,” OpenAI CEO Sam Altman said in an interview in July. "But you can't say for sure, because we might build data centers in space. For example, we could build a large Dyson sphere in the solar system and say, 'Hey, there's no need to build these things on Earth.'"

At the recent US-Saudi Investment Forum, Musk said that the solar energy received by the earth only accounts for one-billionth of the total energy of solar radiation, so entering space is crucial to obtain more energy.

Responding to Musk's point, Salesforce CEO Marc Benioff said space has "continuous solar power, no batteries required" for power and cooling. "The cheapest place to build a data center is in space," he wrote on X.

The latest news is that Beijing plans to build and operate a centralized large-scale data center system with a power exceeding one gigawatt (GW) in a 700-800 km morning and evening orbit to achieve the goal of moving large-scale AI computing power into space. According to the planning plan, the data center system consists of space computing power, relay transmission and ground control subsystems. The construction of the data center is divided into three phases. The first phase is from 2025 to 2027. It will break through key technologies such as energy and heat dissipation, iteratively develop test stars, and build a first-phase computing constellation.

Related articles:

Google CEO: My job will be replaced by AI in the future