Thanks for your reading and for the comment.

You have right in some use cases, it’s wrong for others.

Mine was about IoT devices, water meter, with more than 3 millions of device that provides 35+ metrics every hours. The power of BigQuery to aggregate the data and extract the value was the key. But for low latency, it was an issue. Therefore, we extracted only the aggregated value to put them in Cloud SQL and serve them at low latency, while the data warehouse was still in BigQuery. And it’s far less expensive that Spanner for Tb of immutable data!