BigQuery's on-demand pricing model has dynamic concurrent query limits, with each project running at most a certain number of queries simultaneously by default (typically starting at around 100 slots), and training pipelines may enter a queued state during peak periods.