There isn't enough information to answer that. Some point that can help you:
1) I'd try getting the RAW SQL query and investigating it:
Do you have any replicas? If not, perhaps you can create replicas and segregate a part of this load
3) Besides that, I'm wondering if you're using the right service for this job. Actually, that sounds not, so I suggest you to take a look at GCP BigQuery, Dataflow, and Dataproc services. I think you can improve your architecture using Google's AI and ML services.