xAI's computing monster: the insane numbers behind 2025
The data xAI reveals in its year-end summary shows a simply insane pace of development.
The company expanded its operations from a single work site to five strategic locations in the US.
At the heart of the activity is a formidable computing array comprising over 450,000 GPU units currently operating across several sites, with the stated goal of doubling this quantity by Q2 2026.
This logistical operation requires the work of over 3,000 local technicians, working around the clock, 24/7, to rapidly establish the world's largest supercomputer.
The physical infrastructure required to support such processing power is phenomenal.
Note the data
xAI designed and built a temporary energy system providing hundreds of megawatts of stable power for Macrohard, alongside the installation of over 244,000 miles of fiber optics.
That's right, 244 thousand miles of fiber optics!
This is an intricate network comprising 1,304,608 separate connections, and a water-piping based cooling system extending over 15 miles.
The company's execution capability is also evident in its exceptionally aggressive timelines. From the moment the first mass-produced GB300 was activated, it took the company only 42 days to expand the second-stage cluster to over 100,000 units.
These figures reflect a level of engineering efficiency that is changing the face of the technological race and setting a new standard for production and operational capabilities in the AI industry.
So folks, after reading this, we can give some credit to Musk.