Elon Musk claims that artificial intelligence startup xAI will deploy a 300,000 Nvidia Blackwell B200 GPU data center by next summer.

The Tesla, Twitter/X, and SpaceX CEO provided few details on the supercomputer.

"The xAI 100k H100 liquid-cooled training cluster will be online in a few months," Musk said on his social media platform. "Next big step would probably be ~300k B200s with CX8 networking next summer."

It is important to note that Musk has a history of overpromising on tight deadlines that are then widely missed.

Last week, The Information reported that the 100,000 H100 GPU was targeting a launch of next summer - so this would also represent a dramatic speed-up on the system's timeline.

It is most likely that the company is deploying a much smaller cluster at first, with hopes of expanding it to 100,000 further down the line - a similar approach may occur for the 300,000 system.

Such a supercomputer would also cost tens of billions - more than xAI is believed to have raised. In March, Nvidia CEO Jensen Huang said that B200s cost between $30,000 and $40,000 - taking the lower figure, the GPUs required for Musk's machine would cost $9 billion.

Adding in networking, CPUs, infrastructure, land and facilities costs would push the supercomputer's price tag even higher. And that's before it's turned on and power and water costs come into consideration.

Last month, xAI, which was founded a year ago, closed a $6 billion fundraising round that valued the business at $24bn. That round, which added to a previous $1bn raise, will also have to cover the costs of the 100,000 H100 GPU system, staffing, and current compute needs.

xAI is believed to currently rent around 16,000 Nvidia H100 GPUs from Oracle Cloud.