Is the "Space Data Center" the new battleground for AI? Elon Musk and Jeff Bezos are at odds, and Sam Altman wants a piece of the action too

The trillion - level data center boom in artificial intelligence (AI) is giving rise to a brand - new battlefield: outer space. After a long - standing competition in the rocket and satellite fields, tech billionaires Elon Musk and Jeff Bezos are extending the race to orbital AI data centers, attempting to move the massive computing infrastructure off Earth.

On December 10, according to a report by The Wall Street Journal citing people familiar with the matter, SpaceX, under Musk, plans to use upgraded "Starlink" satellites to carry AI computing loads and is promoting this technology as part of its share sales. This plan could value SpaceX at $800 billion. Meanwhile, a person familiar with the matter said that Jeff Bezos' Blue Origin has also assembled a team and has been conducting research and development on the technologies required for orbital AI data centers for over a year.

The participants in this race are far more than just these two giants. It is reported that Sam Altman, the CEO of OpenAI, has also studied the possibility of acquiring a rocket operator to deploy AI computing power in space. Tech giants such as Google are also actively making arrangements. This emerging trend indicates a potential intersection of the AI and aerospace industries, with the core driving force being to solve the growing energy consumption problem of AI on Earth.

Although supporters are excited about the prospect of utilizing the infinite solar energy in space, sending data centers into orbit faces huge engineering obstacles and questions about cost - effectiveness. Skeptics believe that the technical risks are underestimated, and it is difficult to compete with ground - based facilities in the short term. For investors, this is both an imaginative future opportunity and comes with extremely high uncertainty.

01 Earth Can't Hold AI's Ambitious Plans, Space Becomes a New Energy Solution

The core logic of sending data centers into space is to break free from the physical limitations on Earth, especially the astonishing amount of electricity required for AI model training and inference. Supporters envision that in the future, orbits will be filled with satellites equipped with AI chips, which will operate using the continuous solar energy and transmit the processed data back to Earth.

"The idea of moving resource - intensive infrastructure off Earth has existed for many years, but it requires a reduction in launch and satellite costs. We are approaching this tipping point," said Will Marshall, the CEO of satellite operator Planet Labs.

Senior tech investor Gavin Baker also believes that from a first - principles perspective, space data centers have unparalleled advantages over ground - based facilities in terms of energy and cooling. In space, satellites can receive solar energy that is 30% more intense and up to six times the total amount compared to the Earth's surface, and there is no need to equip expensive energy - storage batteries. At the same time, using the near - absolute - zero environment in space for heat dissipation is almost "free", which can eliminate one of the most complex and costly cooling systems in ground - based data centers. Baker predicts that space data centers may become one of the most important technological breakthroughs in the next three to four years.

02 Giants Compete: Musk's "Starship" vs. Bezos' "New Glenn"

In this space race, launch capabilities are the key to success. Both Musk and Bezos are using their aerospace companies' next - generation heavy - lift rockets to pave the way for the grand vision of space data centers.

According to people familiar with the matter, SpaceX's AI computing technology will be installed on upgraded satellites designed specifically for the "Starship". The "Starship" is a giant, fully reusable rocket that SpaceX is developing, aiming to significantly reduce launch costs. Musk said on social platform X last month that the "Starship" should be able to send about 300 gigawatts (GW) of solar - powered AI satellites into orbit each year, and it may even reach 500 gigawatts.

On the other hand, Bezos' Blue Origin made significant progress this year on its long - developed "New Glenn" rocket project. This partially reusable rocket has a huge fairing and is designed to send a large number of satellites into orbit at once. Bezos said at an event in October that it is reasonable to move data centers into orbit and predicted that its cost will surpass that of ground - based AI infrastructure in 20 years or less.

"It ultimately comes down to launch (capabilities)," said Jonny Dyer, the CEO of satellite company Muon Space. For investors, this means that companies with reliable, low - cost, and high - capacity launch vehicles will have an absolute advantage in this emerging race.

03 New Players Enter the Scene: Altman and Google Also Make Arrangements

In addition to Musk and Bezos, other tech leaders are also showing great interest in orbital data centers, trying to get a share of this potentially huge market.

According to a report by The Wall Street Journal, Sam Altman, the CEO of OpenAI, has explored having his company take over a rocket operator to use its launch vehicles to deploy AI computing power in space. He believes that the huge computing power demand of AI systems may ultimately require a large amount of electricity, making space a better choice.

Meanwhile, Google has reached an agreement with Planet Labs and plans to deploy two test satellites carrying Google's AI chips (tensor processing units) into orbit in early 2027. Google describes this project as one of its "moonshot" projects, highlighting the difficulty of large - scale deployment of satellite data center networks. According to Travis Beals, the Google executive in charge of this project, assuming the use of 100 - kilowatt satellites, 10,000 satellites are needed to replicate the computing power of a gigawatt - level ground - based data center.

In addition, Eric Schmidt, the former CEO of Google who now heads rocket company Relativity Space, has also talked about orbital data centers. IBM's Red Hat software business and Axiom Space launched a data - computing prototype in August this year. Start - ups such as Aetherflux and Starcloud, supported by venture capital, are also formulating plans to compete with large enterprises.

04 Real - World Challenges: Cost, Technology, and Scalability Issues Need to Be Solved

Although the prospects are alluring, space data centers still need to overcome numerous obstacles to turn the concept into reality.

Firstly, there are severe technical challenges. AI chips operating in orbit need to solve a series of problems, such as temperature management, cosmic radiation protection, and how to transmit massive amounts of data back to Earth without significant delay. Any oversight in any link may lead to the failure of the entire system.

Secondly, there are issues of cost and scalability. Skeptics believe that supporters underestimate the technical risks and doubt whether space - based data centers can compete with ground - based facilities in terms of cost, especially when the limitations of ground - based power and other factors are alleviated. As Google executives pointed out, to reach a scale comparable to that on the ground, thousands of satellites need to be launched, which places extremely high requirements on current launch capabilities and cost control.

Investor Gavin Baker also pointed out that for inference tasks, the application prospects of space data centers are clearer, but due to the large scale, it may take a longer time to completely migrate AI model training tasks to space. The road to space - based AI computing power is bound to be long and difficult.

This article is from the WeChat official account "Hard AI", author: Focused on technology R & D. Republished by 36Kr with authorization.