Zero-ETL Architectures for AI Workloads Direct ML Model Access on Cloud-Native OLAP Systems

Abstract

The growing demand for real-time AI insights necessitates a shift from traditional Extract, Transform, Load (ETL) processes toward Zero-ETL architectures. This paper investigates an approach that lets ML models directly connect to cloud-native Online Analytical Processing (OLAP) systems by bypassing traditional data pipelines. Zero-ETL frameworks cut ETL expenses to deliver rapid access to data and provide automatic model updates combined with real-time execution against active operational data. The paper evaluates fundamental architectural elements of serverless computation together with distributed query infrastructure and real-time data extraction schemes that facilitate easy machine learning implementation. The evaluation outlines a comparison between Zero-ETL pipelines and traditional ETL solutions regarding operational complexity together with system throughput and model accuracy. Analysing artificial intelligence workload deployment through three popular cloud-native OLAP platforms BigQuery Snowflake and Redshift demonstrates performance improvement along with deployment advantage. This research addresses consistency and security and model governance issues in Zero-ETL environments through proposed future directions for building scalable intelligent data architectures for enterprises that use AI.