“Long model response times (wall clock time) create a significant bottleneck for AI research and development by slowing down the iteration cycle for experiments and evaluations.”