# Glue vs dbt on AWS — Decision Matrix (July 2026)

| Need | Prefer AWS Glue | Prefer dbt |
| --- | --- | --- |
| Extract from non-SQL / multi-source / streaming | Yes | No |
| Heavy Spark (hundreds of GB–PB, custom Python/Scala) | Yes | No |
| SQL transforms already in Redshift / Athena | Overkill | Yes |
| Analyst-owned models + tests + docs | Weak | Yes |
| Iceberg MERGE / lakehouse ETL at Spark scale | Yes (Glue 5.1) | Partial (dbt-athena / warehouse SQL) |
| Cost unit | DPU-hour (~$0.44/DPU-hr — verify) | Warehouse/Athena query cost (+ optional dbt Cloud) |

## Recommended stack

1. Glue (or Firehose) → raw/staging on S3 / Redshift staging  
2. dbt Core on Redshift or Athena for curated models  
3. Orchestrate Glue → dbt with MWAA or Step Functions  

## Glue runtime pin (July 2026)

- **Glue 5.1** default: Spark 3.5.6, Iceberg 1.10.0, Hudi 1.0.2, Delta 3.3.2  
- **Glue 5.0**: Spark 3.5.4, Iceberg 1.7.1 — still fine; migrate intentionally