Benchmark Dataset for Comparative Analysis of AI Inference Efficiency in Public and Private Clouds

This dataset supports the manuscript titled \u201cBenchmarking AI Inference Efficiency in Public and Private Clouds: An MLPerf-Based Comparative Study,\u201d submitted to IEEE Transactions on Cloud Computing. The dataset includes cleaned and annotated benchmark data from MLPerf Inference Round 5.0,  along with vendor-reported TDP specifications. It also includes processed files used for inferential statistical analysis and residual diagnostics in JASP. Variables include inference throughput, accelerator model, cloud type, TDP (watts), and performance-per-watt efficiency. This data enables replication of t-tests, ANOVA, two-way ANOVA, and regression models assessing the effect of hardware and cloud infrastructure on AI inference efficiency.