For years, the race for artificial intelligence seemed to depend on who designed the best chip. Then came another competition: who trained the most powerful model. Now it has begun a third battle, much quieter and probably more decisive: that of electricity.
China has just presented the Envision Galaxy Campus, a gigantic complex that aspires to become the largest data center for artificial intelligence on the planet. The project plans to deploy up to one million AI accelerators and operate thanks to a direct supply of wind and solar energywithout depending on the traditional power supply model from the electrical network. But the real interest of the project is not only in its size. It’s on to an idea that could forever change the way we build AI infrastructure.
Until just a few years ago, the big bottleneck for AI was getting sufficiently powerful processors. But each new generation of models consumes increasing amounts of electricity. A large data center can require as much power as a medium-sized city, and technology companies are beginning to encounter an unexpected problem: there are places where it is no longer easy to get enough power. The challenge is no longer just to make more computers, it is to power them. Until now, most data centers followed a relatively simple pattern: they were built near large metropolitan areas and then the electrical grid was reinforced to supply them. Envision proposes exactly the opposite.
Instead of bringing electricity to the computers, he wants to bring the computers to where there is excess electricity. That is why the Galaxy Campus will be built in Inner Mongolia, a region with enormous wind and solar resources. There the company intends to directly connect the data center with renewable parks, storage systems and a microgrid designed specifically to power artificial intelligence loads. It is the same philosophy that inspires your initiative Mission Gobi, which aims to develop 5 gigawatts of capacity for green data centers in desert regions before 2030. Which is equivalent to the energy generated by five nuclear reactors… enough for a city of millions of inhabitants.
Although Envision has not revealed the extent of the Galaxy Campus, the energy scale of the project gives an idea of its magnitude. Mission Gobi aims to reach 5 gigawatts of AI capacity, approximately three times the approved power of Galaxy’s Helios campus in Texaswhich occupies almost 9 km². If both complexes maintained a similar density, the Chinese project could extend over several tens of square kilometers.
End of form
At first glance, the desert seems like an odd choice. However, deserts have several advantages that are difficult to match. They have enormous areas of land, receive abundant solar radiationusually have good wind resources and allow large energy facilities to be built far from urban centers. In other words, they offer just what artificial intelligence begins to need more than anything else: abundant energy and room to grow.
The Chinese project also reflects a change in international technological competition. While some American companies seek to secure electricity supply through new gas plants, nuclear energy or small modular reactors, Envision is committed to a different model: creating ecosystems where renewable generation and computing are born together from the beginning.
It’s not just about reducing emissions. It also aims to avoid one of the biggest bottlenecks of modern AI: long waits for new connections to the power grid. Maybe In a few years, another complex will surpass the Galaxy Campus in number of processors or installed power. This happens all the time in the technology industry. But the legacy of this project could be different.
If this model ends up prevailing, the Galaxy Campus will be remembered not only as the largest data center of its time, but as the first to change a rule that seemed immovable: in the era of artificial intelligence, perhaps already It is not the energy that travels to the computers, but the computers that travel to the energy.