IS this for some kind of HPC cluster? What kind of utilization are you at? For an AI company these GPUs are going to be at near 100% utilization 24/7. These kinds of loads destroy hardware quick.