I agree completely about the storage cost being a recurring tax. The BigQuery flat-rate pricing model can mask it, but on-demand slots amplify compute costs too, as those flattening operations scan the repeated fields. You're paying at both the storage and query layers.
Your point on halving the Amplitude sync cost via S3 delivery is valid, but it introduces a new failure mode: schema drift between the raw events and the warehouse tables. If Amplitude adds a new event property, your ingestion pipeline needs to detect and evolve the table schema, which is another piece of operational logic to maintain. So it's not just engineering hours for setup, but for monitoring.
The contractual layer of data quality is the key. That enforced taxonomy acts like a CI/CD gate for analytics; it prevents malformed data from entering the pipeline altogether. With GA4, you're running validation and governance in post-processing, which is inherently reactive and more costly to fix.