Inside the race to build AI data centers in space

Zack Christenson
(Getty Images) 

In the 1966 novel Colossus by British author D.F. Jones, a supercomputer (which goes by the name of Colossus) is given control and decision-making power over the US’s nuclear arsenal – a logical and unemotional computer being better placed, it is assumed, to make unemotional decisions than a human. Eventually, Colossus discovers the existence of a similar supercomputer in the USSR and begins communicating with its Russian counterpart in mathematical languages about technological advances beyond human comprehension.

Frightened by the possibilities this presents, scientists sever the connection – only for Colossus to threaten to launch nuclear weapons if it isn’t reconnected. Colossus and its newfound Russian friend begin monitoring and controlling their human creators in the name of saving humanity from itself.

In January, a Reddit-style web forum, Moltbook, was launched. What’s different about this forum is there are no humans allowed: it’s a place for AI agents to communicate with one another. By February, it was estimated there were 1.6 million agents communicating with one another. Is this Colossus in the making? 

In the AI race for supremacy, it usually comes down to a few main players building AI models – OpenAI and Anthropic perhaps the two biggest startups, along with Google’s DeepMind and X’s Grok. Running an AI company isn’t cheap, which is why there are only a few players. OpenAI and Anthropic have raised a combined $167 billion. It’s estimated that the burn rate for OpenAI will be around $14 billion in 2026, with costs including the huge amounts of data processing power in the form of chips and energy.

Startups have raised millions in venture-capital funding to make orbital data centers a reality

The regulatory hurdles involved in building new data centers aren’t getting any easier to navigate. Local community opposition to building new data centers is on the rise, with at least 20 being blocked in the second quarter of 2025, according to a report from Data Center Watch. In the race for scale, it comes down to money – and how quickly you can deploy that money into data centers. It’s become exceedingly difficult. In the AI arms race, Grok is often considered an afterthought, though many tests show that it’s on a par with its competitors. But in January, Elon Musk announced that xAI, the company that owns X and Grok, the AI large language model that has become ubiquitous on X, was being acquired by one of his other companies, SpaceX.

This may look like a standard merger – bankers and lawyers pushing documents among themselves to increase paper values, bail out shareholders of one side of a company and juice investment bankers’ fees. But as anything with Musk, something more interesting may be going on. The new combined company will now consist of several important subsidiaries – an AI company; Starlink, the satellite firm that provides internet access from space; and, of course, SpaceX, with its capability to launch rockets into space at an incredible rate while running at a cost that gets cheaper all the time. 

This plan brings together several components under one roof to do something to solve AI’s data-center problem – by putting them into space. The concept is relatively straightforward, if not literally space age. The major constraints of data centers on Earth are several-fold. First, energy costs are high, as is even the access to consistent energy sources. In space, data centers can be run fully on solar power, 24/7, with basically zero constraints. Second, cooling is an issue, which is why data centers here need ready sources of water. But the vacuum of space is your cooling source. And the third big issue is regulatory – the great advantage of space being the lack of city-council Karens worrying about how they’ll water their gardens.

With Musk’s combined companies, you have Tesla, which can crank out chips; SpaceX, which can launch the data centers into space; and xAI, which will run the AI models and networks on the data centers. Musk has already started planning, and intends to use the funding from the upcoming SpaceX IPO as his springboard. But his competitors also have their eyes on space. Google has announced Project Suncatcher, an orbital data center that will begin testing next year. Not to be left behind, OpenAI has been exploring deals with startups such as Stoke Space to launch its own data centers. Other startups – Starcloud, Axiom Space and Sophia Space, among others – have raised millions in venture-capital funding to make orbital data centers a reality.

The further we get into the realm of science fiction, the more dystopian the de-accelerators’ predictions. Putting data centers into space, far from the watchful eyes of humans, is yet another thing to worry about. Today, if AI goes wrong, can’t you just unplug it? In the case of Moltbook, that’s true. The agents using the site are, after all, carrying out the directives of their controllers and are bound by the restrictions put on them. But what happens if you can’t unplug them? What if they’re solar-powered and have full access to networked computers on Earth? It starts to feel more like Terminator’s Skynet.

But the more likely – and perhaps more frightening – scenario could be an economic one. Citrini Research recently published a report that sent a ripple through Wall Street, causing a downturn in stocks thought to be most affected by AI. The report’s thesis is that the rapid adoption of AI by 2028 could cause mass white-collar unemployment, which would then create massive debt defaults and falls in consumer spending. “A negative feedback loop with no brake,” the report calls it. Of course, these scenarios don’t usually come to fruition. The loss of stagecoach jobs from the advent of the car didn’t wreck the economy – instead, technology ushered in a new era and more jobs.

In the meantime, xAI and its competitors continue to build the data centers they need here on Earth. In July 2024, xAI broke ground on a new supercomputer housed at a former Electrolux factory in Memphis, Tennessee. It is now thought to be the world’s largest AI supercomputer. And xAI named it Colossus.

Comments