As data centers on the ground become increasingly bulky, consuming huge amounts of electricity and land, an idea once considered science fiction is being seriously considered by the technology world to bring AI data centers into space.
According to leaders in the AI and space industry, it is only a matter of time before these "computing machines" appear on Earth's orbit, even visible in the night sky like artificial planets.
The main driving force comes from the unprecedented energy thirst of AI. In which, Meta, OpenAI, Microsoft, Amazon and many large technology corporations are investing hundreds of billions of USD in data centers. OpenAI alone is said to have committed up to 1.4 trillion USD for related projects.
However, ground data centers are facing many barriers. In many areas, electricity cannot meet the increasing calculation demand. Local people also protest due to concerns about increased electricity bills, water shortages and environmental impacts. In that context, space is considered a potential outlet.
Last November, Google announced the Suncatcher Project, which is the spatial data center expected to test flight from 2027.
Elon Musk believes that in less than 5 years, training AI in space could become the cheapest option. This idea also received the support of Jeff Bezos, Sam Altman and Jensen Huang.
The biggest advantage of data centers in space is almost infinite energy. On orbit, solar panels can access sunlight almost 24/7, without being covered by clouds. In addition, this environment has few legal constraints and no objection.
However, the path to realizing it is still full of thorns. Because the cost of launching materials into space is currently very high, about 8,000 USD per kilogram, even SpaceX's cheapest price is about 2,000 USD/kg.
Meanwhile, each server support can weigh more than a ton. Experts believe that only when the cost is reduced to about 200 USD/kg will this model be economically feasible, which may have to wait until mid-2030s.
Besides cost, technical challenges are also not small. AI chips today are not designed to withstand space radiation.
Cooling in a vacuum environment requires giant cooling systems. Pierre Lionnet, a space economist at Eurospace, believes that many claims about the size of space data centers today are "completely unrealistic".
However, optimists believe that AI and space are converging into new investment opportunities.
“AI is the most attractive field today, while space will be second. And now they are converging,” former SpaceX leader Tom Mueller said.
Despite many controversies and technical barriers, the AI data center in space is gradually shifting from science fiction to a serious direction for the future.
