Announcing General Availability of MinIO AIStor Tables

The first data store to build-in Apache Iceberg™ V3, unifying tables and objects for analytics and AI at scale

We are excited to announce the general, global availability of MinIO AIStor Tables. By unifying Tables and Objects in a single high-performance and Iceberg-native data store, AIStor eliminates structured and unstructured data silos to elegantly power any analytics, AI, and agentic workloads at enterprise scale. This improves agility, allowing data teams to gain secure access more quickly. With AIStor Tables, MinIO is demonstrating Open Table Format (OTF) leadership and is the first in the industry to build the full Apache Iceberg^TM V3 Catalog REST API directly into the data store.

Tables establishes MinIO AIStor as the high performance data store purpose-built to power any analytics and AI workload across on-premises, private, sovereign, and hybrid environments. Key features:

Full Apache Iceberg V3 REST Catalog API: AIStor Tables embeds the Iceberg V3 REST Catalog API directly into the data store as a first-class citizen—simplifying integration, improving security, and accelerating access for analytics teams without external services.
Iceberg Views: a standardized metadata format for logical views across different query engines like Spark, Trino, and Hive, Iceberg Views enables views created in one engine to be easily read and used by others without duplicating storage.
Multi-table Transactions & Version Control: Guaranteed data correctness with atomic, multi-table transactions. Version history improves roll-back, time-series and similar operational and analytical processes at the storage layer.

Our Journey From Tech Preview to GA with Iceberg V3

Following our AIStor Tables tech preview launch, we worked closely with early adopters and thought leaders such as Nomura, Japan’s largest investment bank and brokerage group, to validate real-world requirements.

“By running analytics and AI workloads directly on the same data, MinIO AIStor Tables fundamentally simplifies how we build and operate data pipelines,” said Conor Brennan, Managing Director Risk IT, at Nomura. “It allows us to move faster, reduce operational complexity, and treat all of our data as first-class input for AI.”

AIStor's compatibility with Amazon S3 Tables at our initial launch provided a useful baseline, but this was just a preview of the full scope of innovation we had planned. Our tech preview customers were the first in the world to experience complete Apache Iceberg™ V3 support, including Views, built-in to an object store of any kind whether in the cloud on-on-premises. They also recognized why this deep integration and the capabilities it enables are essential for production analytics and AI.

Let’s look more closely at the important capabilities in Iceberg V3, now built-in to AIStor Tables. For a deeper dive check out this blog on V3 and why it matters.

Deletion Vectors (Replacing Delete Files) to Accelerate Reads: Iceberg V3 introduces deletion vectors, which are compact binary bitmaps that mark deleted rows within data files. Using Roaring Bitmaps encoded in Puffin files, deletion vectors compress efficiently for both sparse and dense delete patterns, significantly improving read performance.
Row Lineage For Fast Incremental Processing: Identifying changed rows has traditionally required expensive scans and comparisons. Iceberg V3 adds row-level metadata, which are unique row IDs and last-modified sequence numbers that enable efficient incremental processing, faster pipelines, and lower compute cost.
New Variant Types for Fast Semi-Structured Data Analysis: Iceberg V3 introduces a Variant type that stores semi-structured data directly in columns using an efficient binary encoding. It supports flexible, nested schemas with better filtering, pushdown, and lower storage overhead than JSON to enable faster analytics on evolving data without sacrificing performance.
Geometry and Geography Types for Efficient Filtering: Geospatial queries on binary columns traditionally require full table scans, driving up time and resource usage. Iceberg V3 adds native geometry (planar) and geography (spherical) types, enabling efficient storage-layer filtering for location-based workloads. The result is faster, more efficient queries and significantly improved performance for geospatial analytics.

AI workloads require both structured tables for features and metrics and unstructured objects for images, audio, video, and documents—traditionally managed in separate systems. AIStor Tables, powered with Iceberg V3, lets structured tables reference unstructured objects by path to create a single discovery layer that simplifies queries, improves performance, and reduces operational complexity. This additionally provides enhanced security through consistent policy application, clearer audit trails and simplified administration by operating within a familiar security framework.

AIStor Tables: Making Unstructured Data Discoverable to AI Agents

AIStor Tables is Just the Beginning

Enterprise analytics and AI workloads demand massive concurrency, predictable latency, and support for mixed workloads at scale. Traditional storage architectures add complexity and performance constraints that limit how efficiently data can be used. Our mission is to simplify this foundation, helping customers increase agility while reducing cost. With AIStor Tables and Apache Iceberg V3, we continue to shorten the path from data to insights.

And we’re just getting started. Modern enterprises are hybrid with significant volumes of data distributed across their on-premises and public clouds, and they often struggle to connect these data sets and deliver fast insights across them. Copying/replicating/moving/ingesting data on-premises into cloud AI and analytics platforms like Databricks takes a lot of time, and drives up costs. We’re laser focused on solving this problem, and you can learn more in this blog where we outline the case for embedding Delta Sharing into object storage. So stay with us as we introduce new capabilities and innovation that leverage AIStor Tables to further simplify operations and accelerate analytics and AI outcomes.

"Apache Iceberg, Iceberg, Apache, the Apache feather logo, and the Apache Iceberg project logo are either registered trademarks or trademarks of The Apache Software Foundation. Copyright © 2025 The Apache Software Foundation, Licensed under the Apache License, Version 2.0."