“The 95% reduction in AI inference cost is unlikely to decrease total customer spending, as workloads will simply migrate to cheaper, more appropriate models.”