Masayoshi Son lately poured chilly water on one in all Elon Musk’s extra futuristic AI concepts: placing knowledge facilities in orbit.
At first look, it feels like one other billionaire disagreement a couple of science-fiction future. Musk sees solar-powered AI satellites as a method to escape Earth’s limits. Son, founding father of SoftBank Group, says the AI race shall be received with compute constructed right here on the bottom, not in house.
I believe the extra attention-grabbing reply is someplace within the center.
Orbital knowledge facilities are usually not as absurd as they sound. Sam Altman has known as the thought “ridiculous,” and at full hyperscale, that skepticism is comprehensible. But dismissing the entire class too rapidly misses why corporations are trying up within the first place.
AI is pushing knowledge middle demand by the roof, communities are pushing again towards new amenities, energy grids are below pressure, and water use has turn into a severe concern in some markets.
Space provides a tempting pitch: ample solar energy, no native zoning fights, no neighborhood noise complaints, no freshwater cooling programs, and much more room to increase.
That is the dream. Like most desires involving limitless solar energy and no zoning conferences, it will get messier as soon as the spreadsheet opens.
The drawback is that transferring a knowledge middle into orbit doesn’t make it cease being a knowledge middle. It nonetheless wants chips. It nonetheless wants energy. It nonetheless wants cooling. It nonetheless wants networking. It nonetheless wants upkeep. It nonetheless wants substitute cycles. It nonetheless has to make financial sense.
And that’s the place the dream will get difficult quick.
Son’s argument is admittedly about timing
Son’s most essential level was not that orbital knowledge facilities can by no means work. His level was that orbital knowledge facilities in all probability will not matter quickly sufficient to resolve the AI race occurring proper now.
That distinction issues.
If the following few years are the crucial window for AI infrastructure, then orbital knowledge facilities are competing towards terrestrial initiatives which are already being financed, permitted, constructed, and linked to energy. Even if SpaceX, Google, Blue Origin, Starcloud, worldwide packages, and others make actual progress, orbit just isn’t going to soak up near-term AI demand at hyperscale.
The AI race is transferring at data-center velocity, not aerospace velocity.
That does not make orbital compute irrelevant. It simply means it is in all probability not the fast escape hatch some folks need it to be.
The pitch is smart on paper
The case for orbital knowledge facilities begins with a really actual drawback on Earth.
AI knowledge facilities want monumental quantities of electrical energy; in truth, US knowledge middle energy consumption is projected to double and even triple by 2028 because of the AI growth. They can require…
