xAI has released new footage of the Colossus 2 supercomputer cluster in Memphis. It is on this supercomputer that Elon Musk’s company plans to scale Grok training, and publications about the project already refer to it as Macrohard.
xAI has released new footage of the Colossus 2 supercomputer cluster in Memphis. It is on this supercomputer that Elon Musk’s company plans to scale Grok training, and publications about the project already refer to it as Macrohard.
This is reported by Kurazh. The article states that the first phase of the cluster already has over 330,000 graphics processors. The publication also claims that the system is capable of producing up to 4.4 trillion operations per second, and its power consumption exceeds 1 GW.
To keep such infrastructure running, Colossus 2 used liquid cooling. Separately, xAI, according to the publication, added 558 Tesla Megapacks with a total capacity of 2.3 GWh. Investments in this energy circuit alone are estimated at more than $545 million. Such a reserve is needed so that the data center can withstand peak loads without failures when training large models.
The name Macrohard attracted special attention. The article presents it as an ironic attack on Microsoft, which is OpenAI’s main partner. In this context, xAI is trying to show not just another data center, but its own infrastructure response to competitors who are developing ChatGPT and other large models on Microsoft’s cloud platforms.
xAI is not stopping there. As part of the second phase, which the publications called Macroharder, the company allegedly plans to add another 220,000 Nvidia GB300s. That is, it is not only about expanding capacity, but also about betting on a newer architecture that should speed up model training and strengthen Grok’s position in competition with ChatGPT and Gemini.
The Colossus 2 story is important because the AI race is increasingly turning into a competition not just of models but also of data centers. The more compute, stable power, and cooling a company has, the faster it can train new versions of its models and scale its services. That’s what xAI is currently investing hundreds of millions of dollars into.
Previously, dev.ua wrote about how Elon Musk launched the first phase of the Colossus supercomputer from xAI — the Memphis Supercluster artificial intelligence cluster is connected to the power grid, and now it consumes up to 300 MW, like a metropolis.
