By the start of the next decade, AI will mostly be a space-based endeavor because operating in orbit will be far more cost-effective, according to CEO Elon Musk.
In a long, far-reaching discussion with cofounder and president John Collison on Thursday, the tech billionaire shared some of his characteristic bold forecasts about how the AI revolution will unfold.
Due to AI’s massive energy needs, limited land for large solar panel arrays—and all the bureaucratic red tape—building new AI data centers will be much cheaper in orbit, where solar panels are five times more efficient than on the ground.
“In 36 months, but probably closer to 30 months, the most economically attractive place to deploy AI will be space,” Musk said. “It will then become absurdly better to be in space. The only real place to scale is space. Once you think about what percentage of the sun’s energy you’re harnessing, you realize you have to go to space. You can’t scale much on Earth.”
He added that the utility industry can’t currently build power plants as quickly as AI requires. On top of that, limits on manufacturing gas and wind turbines fast enough create another bottleneck.
Musk explained that space-grade solar panels are cheaper than terrestrial ones because they don’t need as much glass or reinforcement to withstand weather. Additionally, data center cooling is less of an issue in space.
Given space’s advantages over Earth, he was asked where AI would be in five years.
“Five years from now, I think AI in space will launch every year the total sum of all AI on Earth,” Musk said. “Meaning, my prediction is that in five years, we’ll launch and operate more AI in space annually than the cumulative total on Earth.”
While he’s famous for setting wildly ambitious targets with tight timelines, his next goal was a doozy—even by his standards.
Musk said putting that much AI and solar capacity in space would require about 10,000 launches a year—or one launch every less than an hour each day. SpaceX is the most prolific rocket company, setting a record last year with 165 orbital launches.
He added that SpaceX could hit a 10,000-launch annual pace with 20-30 Starship rockets, though the company will build more—enabling perhaps 20,000-30,000 launches a year.
He pointed out that the airline industry has far higher throughput: global daily flights top 100,000.
Patel then asked if SpaceX would become an AI hyperscaler. “Hyper-hyper,” Musk replied. “If my predictions come true, SpaceX will launch more AI than the cumulative total on Earth from all other sources combined.”
SpaceX is already working toward that goal. Last November, it launched a test satellite with an AI server from startup Starcloud. And last month, it asked the FCC for permission to launch up to 1 million solar-powered satellites designed as data centers.
Of course, there are other challenges to operating in space—like protecting hardware from solar radiation and transmitting enormous amounts of data from orbit to Earth. SpaceX’s Starship rocket is also still in development. But said in a note last month that the challenges of space data centers are more about engineering than physics.
Today’s AI hyperscalers also see the potential and are looking to space. For example, Google’s Project Suncatcher aims to pair solar-powered satellites with AI chips, and a prototype could launch as soon as next year.
OpenAI CEO Sam Altman also considered buying rocket company Stoke Space to put data centers in orbit, the reported in December.
For its part, SpaceX is now an AI company following its merger with Musk’s xAI. This comes as SpaceX is expected to go public this year, raising tens of billions of dollars. During the podcast interview, Musk said public markets have far more money than private ones—possibly 100 times more.
“I just keep tackling the limiting factor,” he added. “Whatever the speed limiting factor is, I’ll address it. If capital is the limit, I’ll solve for capital. If not, I’ll solve for something else.”