xAI is using only 11% of its 550,000 NVIDIA H100 and H200 GPUs.
That means about 60,000 GPUs are actually working.
Meta and Google are each at around 43%.
xAI has grown its hardware very fast. It installed 100,000 H200 GPUs in just 19 days for the Colossus supercluster.
This shows that building huge clusters is only part of the job.
High utilization also needs good software, fast networking, and strong operations.