“If the idea of a mass driver on the moon appeals to you, join xAI,” CEO Elon Musk declared yesterday, following a restructuring that saw a flurry of former executives leave the AI Institute.
This is an interesting hiring strategy after the company merges with Musk’s rocket maker SpaceX and the combined company goes through an expected IPO. You might think that xAI employees would be interested in disrupting traditional software companies using deep learning models, or simply using bad wordplay like “macrohard” to achieve AGI. But instead, Elon goes to the moon.
After outlining plans to build an AI data center in orbit, a key synergy between the two companies, Musk took the idea further. “What if I want to get just over 1 terawatt a year?” Musk asked. “For that, we have to go to the moon…I would like to see a mass driver on the moon launching AI satellites into deep space.”
Beyond the data centers orbiting Earth, Musk says, lies an even bigger computer in deep space. And Musk has said that the best way to accomplish that is to build a city on the moon that will manufacture space computers and then launch them into the solar system in a maglev train.
If that feels like a bit of a stretch, veteran mask watchers know that there are clues as to where the discussion will appear in the video of the all-hands meeting that xAI released to the public. Slides describing the moon base are included at the end of the presentation deck, and during SpaceX pep talks, Musk typically shares a rendering of a SpaceX rocket landing on Mars and speaks passionately about the future of a multiplanetary population.
Notably, the moon base comes just after SpaceX publicly backed away from its long-standing goal of colonizing Mars. Now that xAI is integrated into enterprises, Musk needs a new sci-fi metaphor for the future. In this case, the Kardashev scale is a theoretical measure of galactic civilization devised by the eponymous Soviet astronomer in the 1960s. The idea is to scale up energy use. Early civilizations figured out how to harness all the power sources on the planet and then (hypothetically) built infrastructure to go into space and capture the sun’s energy.
Musk said a moon base could potentially use “perhaps even a fraction of the sun’s energy” to train and operate AI models. “It’s hard to imagine what an intelligence of that size would think,” he told his staff. “But it’s going to be incredibly exciting to see it happen.”
tech crunch event
boston, massachusetts
|
June 23, 2026
In the nine years since Mr. Musk announced plans to explore and colonize Mars, that vision has become an effective recruiting tool for SpaceX. The founding story of Musk’s Mars interests provided a long-term vision to unify the company’s various development efforts and demonstrated the company’s ambitions among other space contractors that have settled into addressing government priorities incrementally. The “Occupy Mars” T-shirt provided a visible symbol of SpaceX’s ambitions.
This is where a hypothetical moon base comes into play. It’s part of a long history of Musk wrapping his company around powerful stories. Mars is home to a million people, and now AI is handling the most interesting future. The creepiness of Mars missions was less evident in Musk’s May 2025 Starship update, which ended with a vision of the now-canceled Tesla Optimus robot crossing Mars.
There was just one problem with SpaceX and Mars. No one wanted to pay to go there. A plan announced in 2016 to repurpose the company’s Dragon spacecraft as a Mars lander was abandoned the following year because the technical challenges were too expensive. And since Mr. Musk unveiled the spacecraft that would become Starship in 2016, its capabilities, originally intended to colonize Mars, have been scaled back to focus on two more lucrative jobs: launching satellites for the Starlink communications network and a $4 billion contract with NASA to land astronauts on the moon.
Unlike multiplanetary civilizations, there may be some logic in having SpaceX buy money-spending AI and social media and build data centers in Earth orbit, especially if predictions of rising demands and costs on the ground come true. Experts suggest it could be possible by the 2030s.
If we were to build a satellite on the moon, many of Musk’s other dreams would have to come true first. Scientists and startups are experimenting with building chips and other precision components in space. But the ability to mass-produce tons of advanced computers on the moon means we live in a universe where it’s dramatically cheaper to go to space, and this is the central requirement of these technologies: getting all the raw materials needed for such an endeavor to the moon, as well as procuring everything needed for “self-sustaining cities.”
In a way, that’s important. This is, uh, a stretch goal. If meme-happy retail investors buy into this argument, they could turn SpaceX stock into the next Tesla. The changes may be uncomfortable for the AI, aerospace and other engineers needed to help Mr. Musk achieve his goals. But this vision is one way to explain what xAI is, other than LLM, which is perhaps best known as perversion. As one of the company’s outgoing executives said on his way home, “Every AI lab makes the exact same thing, and it’s boring.”
Mass-producing a solar-system-sized supercomputer on the moon is a lot of things (I’m about to get an email telling me not to use the word “insane”), but it’s not exactly the same, and it’s not boring.
