The Memphis cluster now has 110,000 GB200 and 440,000 GB300. Three waves of 220,000 GB300 are announced by December, which would bring the whole thing to more than one million chips.
Elon Musk has given the most detailed schedule to date for the expansion of Colossus 2, the xAI computing cluster installed in the Memphis area. According to him, the installation has 110,000 Nvidia GB200 chips and 440,000 GB300 chips. A first wave of an additional 220,000 GB300 is due to be commissioned this week, a second in November, a third by the end of December — this one presented as conditional. In total, the number of installed chips would more than double over the year, exceeding one million units.
Musk has linked the unusual progress step of 110,000 chips not to the supply of silicon but to the network: this figure would correspond to the number of fiber optic cables that can be connected to a central switch. This clarification is a reminder that the practical limit of these very large clusters is shifting towards interconnection and power supply rather than the availability of accelerators.
The announcement is part of the computational race between xAI and OpenAI, Google and Anthropic, the company seeking to make up for a capacity gap. At the same time, it is pursuing a Nvidia chip rental scheme valued at around twenty billion dollars.
Read the full article on the original source
Go to the source →