MinIO,a pure-play exascale object store for AI data, has announced the addition of native Apache Iceberg support to its flagship AIStor offering through a powerful new Tables feature. By integrating the Iceberg Catalog API directly into MinIO AIStor, MinIO will help enterprises unlock more value from AI on their own terms with an AI Data Store that brings together structured and unstructured data under a single, open standard.
Apache Iceberg has emerged as the de facto standard for enterprise AI and analytics, resolving years of format fragmentation. It had traditionally been considered primarily a tool for analytics on structured data. Now with AIStor’s native Iceberg implementation, enterprises can now unify all their structured and unstructured data – including tables, transactions, images, and audio – into a single, coherent fabric. The resulting foundation strengthens the efficacy of AI workloads from analytics to agentic AI, because AI can now act on all enterprise knowledge, not just a subset.
“Apache Iceberg is a high-performance open table format (OTF) for large analytical datasets, primarily in data lakes,” said Jason Nadeau, VP of Product Marketing at MinIO. “It manages data files as tables, providing advanced features such as schema evolution, time travel, and efficient query planning through a metadata layer.
“There are other OTFs, like Delta Lake and Apache Hudi, but in recent years, Apache Iceberg has emerged as the clear winner and industry standard for organizing large-scale enterprise data,” Nadeau continued. “This is because Iceberg was designed for modern data lakes and data stores, ensuring transactional consistency at scale, without the legacy baggage of filesystems. For enterprises that want a future-proof foundation for AI and analytics, it starts with Iceberg.”
AIStor Tables is a native implementation of the Iceberg Catalogue REST API, embedded directly into MinIO AIStor. Enterprises sit on massive volumes of data, but turning it into AI value has been costly and complex,” Nadeau stated. “Iceberg is the new standard, yet managed cloud services drive up expenses and on-prem solutions with external catalogue services that create operational headaches. By embedding Iceberg natively, AIStor Tables makes data fluent, fast, and AI-ready, removing layers of cost and complexity from data and infrastructure management while ensuring consistency, security, and scale.
“AIStor Tables enables customers to unify all their structured and unstructured data – whether tables, transactions, images, audio and more – into a single, coherent fabric. This foundation strengthens the efficacy of AI workloads from analytics to agentic AI, because AI can now act on all enterprise knowledge, not just a subset. By embedding Iceberg natively, AIStor Tables makes data fluent, fast, and AI-ready, removing layers of cost and complexity from data and infrastructure management while ensuring consistency, security, and scale.
“Imagine all of your enterprise data is a book,” Nadeau commented. “The Iceberg Catalogue is the bookmark that lets you find where you left off reading. You have the freedom to move around your bookmark and return to an earlier chapter you preferred. Adding new pages is easy, as this book scales effortlessly. All of your reading material; unstructured modalities like images, and structured material like Apache Iceberg tables are available in the same place. One cover, one security umbrella, one interface. All your data is findable, manageable and curatable.”
The integration was done in-house by MinIO.
“It was done with a unique approach that mirrors how data teams naturally think about organizing information,” Nadeau said. “Most Iceberg implementations are bolted on and require a separate metadata database, which puts a limit on how much it can scale. Our implementation is embedded directly into MinIO AIStor so tables become first-class citizens alongside objects. It operates like a stateless environment with no need for an external metadata database or a separate catalog service. There is just one unified platform designed for exascale AI workloads.”
Just as MinIO brought the Amazon S3 experience to every data center, MinIO is now doing the same for Iceberg with AIStor Tables – making the enterprise AI standard available at scale, on premises. MinIO designed the Tables feature to address the challenges exascale AI customers face today in building Iceberg-powered workloads as well as repatriating such workloads from the cloud to cut costs while strengthening privacy, sovereignty, and control.
“Data is the currency of AI, but enterprises are struggling with its enormous scale and variation,” said AB Periasamy, co-founder and co-CEO of MinIO. “Iceberg is the clear standard for enterprise AI data. The challenge is that most on-prem implementations make it harder than it needs to be, requiring separate catalogue databases and extra layers of infrastructure that add cost and operational risk. By building Iceberg directly into AIStor, we take away that complexity and give enterprises a simple, scalable foundation for AI. This not only lowers costs and speeds progress, but also ensures AI can reach its full potential because all data is AI data.”
The ability to unite structured and unstructured data is a major win for MinIO.
“As data became a bigger part of our day-to-day, data lakehouses needed database-grade reliability at massive scale,” Nadeau stated. “Transactions in the modern era are not row updates but long-running, distributed operations involving thousands of large and small objects in Parquet, ORC, or Avro-like formats, often across multiple engines. Classic database engines could not handle this. Apache Iceberg solved the problem with a metadata layer and a set of APIs that provide ACID transactions, versioned snapshots, schema evolution, and O(1) metadata operations. Proven at Netflix scale and adopted across the industry, Iceberg became the foundation for all enterprise data. Iceberg has already become the standard for analytics. Its next role is even more important: to become the foundation for enterprise AI data. By unifying structured and unstructured data into a single, governable model, Iceberg allows enterprises to prepare for agentic AI without tearing apart their existing infrastructure.”
So why has this been so hard to do?
“It’s the scale of data that complicates this type of integration,” Nadeau said. “Typically, such integrations require separate catalogue databases and extra layers of infrastructure that add cost and operational risk. Implementing the Iceberg Catalogue API typically requires the deployment and management of separate external catalogue services. This adds to operational complexity, like setting up a JDBC metastore with a dedicated PostgreSQL database, which is required to enable the core functionalities of Iceberg. Second, organizations must engage with a separate security model for this external catalog, differing from the native identity and access management (IAM) controls applied to the underlying object storage data. This situation creates a significant governance implementation gap, resulting in a two-world problem for security, which transforms initial enthusiasm into a complex setup and ongoing administrative burden.
“By implementing this standardized subset of the API directly within the object storage layer, MinIO achieves an optimal balance,” Nadeau emphasized. “This approach delivers the most critical day-to-day functionality while minimizing the operational and security overhead associated with deploying a separate, full-featured catalogue service. The integrated core approach significantly lowers the barrier to entry, enabling teams to become productive with Iceberg quickly.”
Competition in this area has been limited.
“AWS has also built Iceberg directly into AWS S3 via S3 Tables, but managed cloud services drive up expenses,” Nadeau stated. “Enterprises building on-prem for lower cost and more control have to deal with external catalogue services, which create operational headaches. So enterprises have been left with a painful choice: operate three separate subsystems for every Iceberg workload (object store + catalog service + metadata DB) or stay locked in the cloud. This is the first and only object-native on-premises object store with Iceberg built-in. MinIO is uniquely positioned to do this because it was built to scale just like AWS S3, but for on-premises and private environments.
This unlocks key breakthroughs for scalable Enterprise AI, including a unified view of the business that comes from bringing together structured and unstructured data into a single Iceberg-based foundation, ensuring AI has access to all enterprise knowledge. It is also a seamless ecosystem fit, since Tables is built on open standards, and thus AIStor Tables works out of the box with existing tools and query engines including Spark, Trino, Dremio, and Starburst. Native design eliminates separate catalogue service infrastructure, security models, and management overhead of traditional Iceberg deployments, so enterprise AI scales more efficiently. Tables also runs in the private cloud to keep sensitive data secure and compliant, while reducing dependency on hyperscalers and lowering long term AI infrastructure costs.
The introduction of AIStor Tables reflects MinIO’s broader evolution from object storage to AI Data Store pioneer. In the new AI economy, data strategy defines leadership. MinIO gives enterprises the tools to unify, govern, and accelerate their data fabric, on premises, at any scale. This transforms AIStor from an object store into the foundational AI Data Store that agentic AI demands, unifying years of siloed data into a single, governable platform.
AIStor Tables is currently in tech preview but Nadeau said customers will see the value right away. He also said that it won’t cost them anything extra.
Nadeau also pointed out the channel implications.
“Iceberg table integration opens the door for partners to lead with thought leadership around data modernization and AI readiness,” he stated. “It also creates a service opportunity to support customers migrating from existing Iceberg or external catalogue implementations to AIStor.