The real reason why Musk, Bezos and Pichai want to build data centers in space: bypass regulation

The construction of data centers is proliferating so much that although the largest in the world They are in Kolos (Norway), in The Cidatel (United States) and China, you can find them now even in Botorritain the province of Zaragoza. The limit is the sky. Or well, not even that: because Silicon Valley has been put between eyebrows set up data centers in space.

And the main big tech companies are making moves to achieve this. Former Google CEO Eric Schmidt bought rocket company Relativity Space with that objective. Nvidia has supported the startup Starcloud in its project to launch the first NVIDIA H100 GPU into space a few weeks ago and Elon Musk has even condensed how he would do it in a tweet: “It will be enough to scale the Starlink V3 satellites, which have high-speed laser links.” He when Jeff Bezos slipped it in a prediction at the Italian Tech Week: We will see “giant training clusters” of AI in orbit in the next 10 to 20 years.

The moon is a gift from the universe

The next question would be “why?”. The reality is that there is no shortage of reasons. AI is a real energy guzzler and as demand does not stop growingspace offers a couple of differential advantages over Earth: almost unlimited energy and free cooling. On the one hand, in space we have a sun-synchronous orbit where solar panels receive energy almost continuously. On the other hand, you can install a radiator so large that the space functions as a kind of ‘infinite heat sink at -270°C’. The enormous amounts of water essential for cooling on Earth would not be needed.

Let’s face it, today there are no plans to have data centers in space. But not too far away: University of Central Florida research professor and former NASA member Phil Metzger esteem that perhaps within a decade it could be economically viable. But its viability is so clear that it considers that taking AI servers into space are “the first real business case that will give way to many more“in the face of a future human migration beyond Earth.

So for now, they try it on Earth. Consequence: that Donald Trump declare an energy emergency due to the enormous electricity demand expected for the coming years. As the power grid catches up (or tries to), AI companies have decided to move from a passive to a proactive position: Meta is going to become an electricity marketer. xAI by Elon Musk is using gas turbines as energy sources temporary. OpenAI is pushing to the United States government to lend a hand to electricity companies to add 100 gigawatts per year.

That figure doesn’t say much, but it is astronomical: what OpenAI is asking for is that The United States built almost an entire Spain (around 145 GWh considering the 129 GW consolidated at the end of 2024 plus the solar and wind deployment of 2025) every year and a half in terms of infrastructure.

AI is growing faster than electrical bureaucracy is advancing

How could the Trump Administration help? With the eternal bureaucracy. Because on Earth they face great technical challenges, but they also face a legislative wall. To have more energy, the simplest and most immediate step is to build new power plants, but that means successfully going through the tangle of procedures that slow down the process. There is only one small problem: that in the United States depending on technology, it can take five to ten years… if you’re lucky.

Interconnection to the grid alone can take six years, successfully overcoming an interconnection queue with more than 2,000 GW in projects who are already in line. Then, up to four years of federal and environmental permits to end in another couple of years for state and local licenses that must come to fruition. ‘Permit Stack’ they call it.

And the journey does not end here: they must also avoid andthe citizen movementNot in my backyard‘ (not in my backyard, kind of like “yes, but not in my house”), which has already backed down the Battle Born Solar Project (Nevada), which was going to be the largest solar plant in the United States, or Danskammer gas station (New York), among others. This can delay the operation even further as rights of way must be negotiated with individual owners who may refuse, going through the courts again. The never ending story.

To avoid processes NIMBY that last fifteen years or more, companies like OpenAI or Microsoft are buying plants that already exist, such as Three Mile Island, which is going to reopen only for Microsoftinstead of trying to build new ones from scratch. Amazon has also signed infrastructure that is already on the network like the Talen Energy Campus and it has partnered with Dominion Energy and X-energy to develop mini reactors (SMR). SMRs are also Google’s solution, in this case thanks to an agreement with Kairos Power.

Everything is to avoid that tangle of ‘Permit stack’ procedures that in practice and according to estimates, makes it is faster to opt for the space route to build a power plant on the old, familiar Earth. At the end of the day for AI companies “The moon is a gift from the universe”, as already Jeff Bezos glimpsed.

In Xataka | Musk has created the perfect circle: Tesla’s megabatteries power the AI ​​that will define its next cars

In Xataka | Researchers have dismantled the batteries of Tesla and BYD. You already know which one performs better and is much cheaper.

Cover | İsmail Enes Ayhan and NASA