Musk looks to space as only way to meet AI compute demand by 2028

The world’s first trillionaire, Elon Musk, has said that orbital compute is the only way to continue scaling AI by 2029. He made the statement in an X post. He pointed to power shortages and delays in granting permits on land as the main reasons.  

For companies thinking of where to build the coming generation of AI infrastructure, Musk’s statement reads more like a hard ultimatum than a pitch in itself. 

Musk argues that group-based AI compute will become impossible by the end of this decade. It is relevant because presently, regulators have received filings for a million orbiting machines.

Musk puts a due date on when space becomes a must

In his X post, Musk talked of orbital compute as the only alternative to scale AI infrastructure. He blamed “power availability& permitting problems on land.” This marks a shift in Musk’s assessment of the importance of space data centers in the AI race. He previously saw space as an expansion route. Now, it has become a criterion for AI to keep growing.  

Power is a major factor in Musk’s argument, as AI data centers may take up to 17% of US electricity by 2030. 

Fossil-fuel plants are under construction just to meet those needs. Older fossil-fuel plants are worn out, and it’ll require lots of already scarce water supply to cool the machines. However, in space, solar energy is free to obtain and, more importantly, heat can be disposed of into space.

SpaceX, Blue Origin and Google have made applications for orbit

The number of filings for orbit machines shows the idea has gained so much traction. SpaceX (NASDAQ: SPCX) applied to the Federal Communications Commission to build a data-center system of 1 million satellites in space. 

The company’s 2026 filing shows AI compute satellites could become active by 2028. Such a projection is dependent on Starship’s projected ability to haul 100 metric tons to low-Earth orbit.

Blue Origin will not be left out of the fun, as Jeff Bezos’ space firm applied with the FCC for a projected network of 51.600 satellites in March. Project Sunrise, as it’s called, will move energy- and water-hungry compute into orbit. 

Google (NASDAQ: GOOG) is already looking into orbit as well. The company is running Project Suncatcher on its own TPU chips, with two prototype spacecraft planned for a 2027 launch in collaboration with Planet. 

Starcloud, an AI startup, has a 60,000-satellite constellation in the works. It flew the very first Nvidia H100 GPU in orbit last November. It test-ran AI workloads with the chips.

The physics and economics are yet to match up

Musk’s 2029 ultimatum naturally has certain obvious obstacles. A single one-megawatt orbital system requires ~5,640 square meters of solar panels and 2,500 square meters of radiators, without adding the spacecraft. 

The proposed launch would be 3.4 and 13.5 times cheaper than a public Falcon 9 dedicated launch benchmark. All these are findings from a 2026 technical study. 

There is also the risk of AI processors becoming obsolete. AI processors run for a year or two years while satellites run on a 5-7 year cycle. This may mean an orbital machine loses value while the satellites are still valuable.

Don’t just read crypto news. Understand it. Subscribe to our newsletter. It's free.