Creating a unique unigram dataset for each of those one million homepages took just a single SQL query in BigQuery, processing all half trillion words in just over 28 minutes to yield just over 1. ...