Amazon Redshift vs Google BigQuery: $0.375/RPU-Hour Compute vs $6.25/TiB Scanned & JSON TCO

Amazon Redshift Serverless and Google BigQuery approach warehouse monetization from opposite ends of the execution engine. Redshift Serverless bills strictly for compute capacity consumed at $0.375 per RPU-hour in us-east-1 ($0.36 in us-east-2 and us-west-2), metered per second after a 60-second minimum whenever a workgroup wakes up. Whether a 32-RPU workgroup ($12.00/hour) scans 100 GB or 15 TB during that active hour, the compute invoice remains $12.00—and querying external S3 parquet files via Redshift Spectrum incurs zero per-terabyte scan surcharges under Serverless.

Google BigQuery defaults to On-Demand byte-scanning at $6.25 per TiB after a recurring 1 TiB monthly free tier, making it unbeatable for sporadic workloads that sit idle 90% of the day. However, once an organization's scan velocity exceeds 1.92 TiB per active hour ($12.00 / $6.25), a 32-RPU Redshift Serverless workgroup becomes cheaper per hour. Inside BigQuery's own ecosystem, the mathematical break-even between On-Demand ($6.25/TiB) and BigQuery Editions occurs at 156.25 slot-hours per TiB on Standard ($0.04/slot-hr) and 104.17 slot-hours per TiB on Enterprise ($0.06/slot-hr)—meaning a 100-slot Enterprise baseline ($4,380/month) pays for itself once monthly scans pass 700.8 TiB.

Handling semi-structured event streams exposes major architectural and billing differences between both platforms. Redshift's SUPER data type ingests schemaless JSON documents up to 16 MB per row into a post-parse binary representation and dynamically subcolumnarizes frequently accessed attributes into columnar storage blocks. Because Redshift bills by RPU execution time, analysts can run PartiQL unnesting queries across billions of SUPER rows without triggering byte-scan surcharges.

BigQuery's native JSON data type accommodates much larger payloads—up to 100 MB per row and 500 nesting levels—and automatically subcolumnarizes fields on write so the storage engine only reads the targeted sub-fields from disk. However, under BigQuery On-Demand billing, referencing a single nested JSON path still bills for the entire logical UTF-8 byte footprint of that JSON column. To avoid runaway scan bills on heavy JSON tables in BigQuery, teams must transition to BigQuery Editions (slot-hour billing) paired with Physical Storage billing ($0.04/GiB active, $0.02/GiB long-term), which typically compresses repetitive JSON keys by 60% to 80%.

Each warehouse carries a distinct operational cost trap if left unmonitored. In Redshift Serverless, the primary trap is the 60-second minimum billing floor combined with uncapped Max RPU settings: if an automated dbt job or monitoring script fires a 2-second query every 3 minutes, the workgroup stays perpetually awake or bills a full 60 seconds per wake-up cycle, turning a light workload into a 24/7 cluster bill. Engineers must configure strict `max_capacity` ceilings and daily RPU usage limits to cap monthly spend.

In BigQuery, the classic trap is unpartitioned BI dashboarding on On-Demand pricing, where Looker or Tableau users trigger full-table scans at $6.25/TiB without a `Maximum Bytes Billed` guard rail. On the storage side, Redshift Managed Storage charges a flat $0.024/GB-month for compressed data on S3 with no automatic cold-tier discount, whereas BigQuery automatically drops unedited partitions by 30% to 50% after 90 days ($0.023 to $0.016/GiB on Logical, or $0.04 to $0.02/GiB on Physical compressed storage).