Elon Musk believes the AI boom may have roughly three years left before Earth itself becomes the bottleneck.
“Orbital compute will be the only way to scale AI probably sometime in 2029 due to power availability & permitting problems on land,” the SpaceX (NASDAQ:SPCX) CEO said in a post on X Friday.
Why Musk Is Circling 2029
JPMorgan recently modeled SpaceX expanding its terrestrial compute capacity to roughly 9 gigawatts by 2029, about four times Hoover Dam’s nameplate generating capacity.
“Beyond 2029, we expect SpaceX to pivot to orbital compute for incremental capacity additions,” the brokerage said. That’s strikingly close to the point Musk now says Earth could become the bottleneck.
Terrestrial AI campuses require enormous amounts of electricity, transmission, land and approvals. Musk’s argument isn’t that data centers stop working on Earth in 2029. It’s that building the next gigawatt may eventually become easier than finding somewhere to plug it in.
Space Data Centers Aren’t as Crazy as They Sound
Putting thousands of expensive AI chips into orbit still sounds absurd to many critics. Peter Zeihan called the idea "so many levels of dumb."
But Patrick Bowen, CEO of photonic chip startup Neurophos, which raised a $110 million Series A led by Bill Gates’ Gates Frontier, previously told Benzinga there is “no single engineering obstacle that can’t be overcome for this application.”
Bowen says the math could improve dramatically with the energy-efficient photonic chips Neurophos is developing: less power means less waste heat, smaller radiators and ultimately less mass to launch.
“We believe that Elon Musk is in a unique position to make compute in orbit viable,” Bowen said.
The Catch Is Starship
For SpaceX, the economics depend heavily on Starship becoming rapidly reusable, sharply reducing launch costs and flying at high cadence.
SpaceX aims to begin orbital AI compute demonstrations by late 2027. But proving the concept is very different from putting enough compute into orbit to rival a data center packed with hundreds of thousands of accelerators.
Traders Still Think Musk Is Early
Kalshi traders currently give a 1-megawatt orbital data center just a 15% chance of going live before January 2029, rising to 39% before 2035.
For Musk, the real test isn’t whether AI chips can work in space. It’s whether Starship can make launching them cheaper and faster than building the next data center on Earth.
Image: Shutterstock
Kalshi and Benzinga have an existing data collaboration agreement.
Login to comment