Elon Musk: ‘In 30 Months, the Cheapest Place for AI Will Be Space’

As you know by now, every tech company on Earth is racing to build bigger AI data centers. Well, Elon Musk thinks they’re all looking in the wrong direction. Literally.

In a three-hour interview on John Collison’s Cheeky Pint podcast (with Dwarkesh Patel co-hosting), Musk laid out why he merged SpaceX with xAI, and the core argument is deceptively simple: we’re running out of electricity.

Here’s the crux of it:

• The entire US uses about half a terawatt of power. Scaling AI to where it needs to go would mean doubling that.
• Meanwhile, chip production and demand is growing exponentially while power generation outside China is basically flat. Natural gas is an ideal training data source, but gas turbine manufacturers are backordered through 2030.
• “Those who have lived in software land,” Musk said, “don’t realize they’re about to have a hard lesson in hardware.”

His solution? Put the data centers in orbit, of course! Why?

1. Solar panels in space produce ~5x more power than on the ground (no atmosphere, no clouds, no nighttime; batteries not required).
2. Musk’s estimate: space-based solar is effectively 10x cheaper than terrestrial.
3. “You can mark my words. In 36 months, probably closer to 30 months, the most economically compelling place to put AI will be space.”

Well, here’s the thing about marking Elon’s words… they’re a bit, how to put it… oh, I know:

There’s a reason for that, though. As he says in the interview, as an executive, he pushes as hard as he can on the limiting factor to speed. It SHOULD be possible in 36 months… but there’s a lot that can go wrong, of course.

Here’s what else the interview covered:

• TeraFab: Musk is planning a chip factory producing 100M+ advanced chips per year. Current fabs are fully booked, and creating a new one is really hard. If power is the limiting factor right now, chips will be the limiting factor in 5 years.
• Optimus robots: Building a humanoid hand required custom actuators, motors, gears, and sensors from scratch. No existing supply chain exists.
• China’s lead: It produces 3x the US’ electricity output and accounts for 98% of global gallium refining (important for high-end electronics). Musk’s blunt assessment: “We can’t win on the human front, but we might have a shot on the robot front.”

Here’s why this matters

Musk’s companies are now a vertically integrated bet on AI scaling. SpaceX launches the satellites, Tesla makes the solar panels and robots, xAI builds the AI. It’s one system engineered to scale past Earth’s physical limits and capture as much power from the sun as possible to rise up the Kardashev scale:

• ICYMI: Proposed in 1964 by Nikolai Kardashev, the Kardashev scale measures a civilization’s technological advancement based on the total amount of energy it can harness and utilize, ranking them into three categories:
• Type I = planetary (harnesses all energy reaching its planet from its host star, enabling total control of its world’s resources).
• Type II = stellar (harnesses the entire energy output of its star, capturing billions of times more power than a Type I).
• Type III = galactic (harnesses the energy of an entire galaxy by tapping into billions of stars and massive black holes).

Five years from now, Musk predicts SpaceX will launch more AI compute annually than the cumulative total on Earth. Whether or not you believe the timeline, the underlying constraint is real: chips are useless without power, and we’re running out of places to plug them in. If we can power them from space, well, then the only limiting factor will be producing them.

Read the full transcript on Dwarkesh’s blog here.

Editor’s note: This content originally ran in the newsletter of our sister publication, The Neuron. To read more from The Neuron, sign up for its newsletter here.

The post Elon Musk: ‘In 30 Months, the Cheapest Place for AI Will Be Space’ appeared first on eWEEK.