Adopting new hardware will require modifying security-critical code
Another concern is that AI companies (or the AI company) will rapidly buy a bunch of existing hardware (GPUs, other accelerators, etc.) during the singularity, and handling this hardware will require many infrastructure changes in a short period of time. New infrastructure might be needed to handle highly heterogeneous clusters built out of a bunch of different GPUs/CPUs/etc. (potentially including gaming GPUs) bought in a hurry. AI companies might buy the hardware of other AI companies, and it might be non-trivial to adapt the hardware to the other setup.
Another concern is that AI companies (or the AI company) will rapidly buy a bunch of existing hardware (GPUs, other accelerators, etc.) during the singularity, and handling this hardware will require many infrastructure changes in a short period of time. New infrastructure might be needed to handle highly heterogeneous clusters built out of a bunch of different GPUs/CPUs/etc. (potentially including gaming GPUs) bought in a hurry. AI companies might buy the hardware of other AI companies, and it might be non-trivial to adapt the hardware to the other setup.