Redazione RHC : 22 September 2025 06:54
Elon Musk shared a photo of the complex housing Colossus II, the world’s first gigawatt-scale AI cluster. Elon Musk has previously stated that the Colossus 2 supercomputer, developed by xAI, will become the world’s first gigawatt-scale computing center for AI training.
The system, designed to power the Grok chatbot and other AI projects, will use 550,000 Nvidia GPUs, including GB200 and GB300 chips .
“Our goal is to get as close to the absolute truth as possible,” Elon Musk said, emphasizing that Grok should provide the most accurate answers possible.
Elon Musk previously announced that xAI was currently training its Grok model on 230,000 GPUs, 30,000 of which are the latest Nvidia GB200. All this computing power is concentrated in a supercomputer called Colossus 1.
The race for supercomputers marks a turning point in the global competition for artificial intelligence. Following Elon Musk’s announcement regarding Colossus 1 , it’s now clear that the game is being played on scale and the ability to support unprecedented computing infrastructures. Each new step forward is no longer just about technological innovation, but the very possibility of maintaining such energy-intensive systems.
In this regard, power plants are becoming the new strategic field. AI computing networks require impressive amounts of energy, so much so that some industry players are looking to nuclear power plants as the only resource capable of guaranteeing continuity and stability. It’s not just a matter of finding new sources, but of redefining the relationship between computational power and energy availability, with the risks of concentration and geopolitical imbalances that could affect future global economic balances.
Against this backdrop, Microsoft has made an unprecedented move by securing power for two nuclear power plants to support its AI ambitions. This move highlights how energy has become an integral part of the technological value chain: without a constant power supply, supercomputers and AI models cannot exist. The challenge is no longer simply to create more sophisticated models, but to sustain them over the long term, in a race where the winner will be the one who can combine innovation, computation, and energy capacity.
Redazione