MinIO

Overview

MinIO is S3-compatible object storage you run yourself, and anyone choosing it today needs to know one thing before anything else: the open-source community edition is over.

The GitHub repository now opens with a notice that it is no longer maintained, and it is archived and read-only. MinIO's own pricing FAQ states the AGPL edition's codebase is frozen, with no new features, bug fixes or security patches. The last community release was in October 2025.

Features had already been moved out before that. The May 2025 release deprecated the embedded web console and removed LDAP and OIDC logins, directing users to the commercial product.

The actively developed product is AIStor, described as high-performance, Iceberg-native and fully S3-compatible, deployable across edge, core and cloud. MinIO is careful to say the company has not discontinued anything, only the community codebase.

Note what "AIStor Free" means: it is a free commercial licence for a single-node deployment with redistribution prohibited, not an open-source edition.

Technically it remains strong. Objects reach about 48 PiB through multipart, a soft limit of 500,000 buckets, and no hard limit on servers, drives or objects per bucket.

Performance comes from the software design rather than managed infrastructure: SIMD-accelerated erasure coding, zero- copy buffers, direct I/O by default, and a sharded metadata cache tuned for Trino and Spark.

The lakehouse feature is genuinely distinctive. AIStor Tables embeds a full Iceberg V3 REST catalog into the storage layer itself, removing the need for Hive, Glue, Nessie or a catalogue database.

Compliance credentials are published and unusually specific for self-hosted software: FIPS 140-3, SOC 2, ISO 27001, SEC 17a-4(f) and FINRA 4511.

Pricing is capacity-based and quote-only, charged on usable capacity after erasure-coding overhead rather than raw disk, which is the right basis but gives you nothing to budget against without a conversation.

Key features and capabilities

The same headings are used for every object storage entry, so two tools can be read side by side.

Storage model

• S3-compatible buckets and objects, with warehouse buckets and table namespaces added by AIStor Tables
• Maximum object size around 48 PiB through multipart, and 5 TiB in a single upload
• A soft limit of 500,000 buckets, with no hard limit on servers, drives or objects per bucket
• Versioning defaults to 10,000 versions per object and is configurable

Performance and scale

• Multipart uploads of up to 10,000 parts, with 64 MiB to 512 MiB parts cited as optimal
• SIMD-accelerated erasure coding, zero-copy buffer pools and direct I/O by default
• A 4,096-shard metadata cache, tuned for Trino and Spark, with sub-two-millisecond collection pauses
• Throughput ceilings are not published because they depend on your own hardware

Storage classes and lifecycle

• Lifecycle management and tiering to external targets are part of the data management feature set
• Storage classes are erasure-coding parity settings rather than priced tiers
• Read quorum is the drive count minus parity shards; write quorum one higher
• Minimum durations and retrieval fees do not apply, since nothing is metered per GB

Security and governance

• IAM policies, roles, temporary credentials and OIDC federation
• LDAP and OIDC login are commercial features, having been removed from the community build in May 2025
• Server-side encryption with a dedicated key management server
• Object locking, legal holds, object immutability and ransomware protection
• Published credentials, FIPS 140-3, SOC 2, ISO 27001, SEC 17a-4(f) and FINRA 4511, plus audit trails and Prometheus metrics

Analytics features

• AIStor embeds a fully compliant Iceberg REST catalog in the storage layer, with no external catalogue database
• Spark, Trino, PyIceberg and Flink connect through the standard Iceberg REST catalog API
• Iceberg capabilities include atomic commits, time travel, and schema and partition evolution
• Table sharing to external consumers through the Delta Sharing protocol

Ecosystem

• Full AWS S3 API compatibility, including policies, versioning, lifecycle and multipart, with no code changes
• Both path-style and virtual-host-style requests, with per-bucket domains and SNI certificate selection
• Additional protocols, S3 Express, SFTP, Iceberg catalog, Delta Sharing and MCP
• Components include the server, a CLI client, an SDK, a key management server, a Kubernetes CSI driver and a benchmarking tool

Where it runs and what it costs

• Self-hosted across edge, core and cloud, with no vendor regions and no managed service
• One server with two drives works for testing; there is no hard limit on servers or drives
• Appliance route through certified Dell and Supermicro hardware sold via Arrow, from 1.1 PiB usable
• Reference hardware is NVMe drives of 30 TB or more, 100 GbE networking and 256 GB of memory per node

Pricing

Price on requestQuote only; AIStor Free for a single node

No monetary figures are published. AIStor Free costs nothing for a full-featured single-node deployment with community support, but prohibits redistribution and is not open source. Enterprise Lite covers multi-node deployments below 400 TiB, with an optional under-five-day support commitment; above that you move to Enterprise, which adds 24/7 direct-to-engineer support with a four-hour commitment. Charging is based on usable capacity after erasure-coding overhead, not raw disk, with volume discounts. A 60-day enterprise trial licence is available.

Vendor pricing page →

Demos and videos

• Getting started with MinIO AIStor Product demo · YouTube
• MinIO introduces AIStor Tables Webinar or talk · YouTube

About MinIO

MinIO, Inc. was founded in 2014 by Garima Kapoor, Anand Babu Periasamy and Harshavardhana, and is based in Redwood City, California, with Kapoor and Periasamy as co-chief executives. Its licensing position changed materially in 2025 and 2026: AIStor is commercially licensed, the frozen community codebase remains under AGPL v3 but archived, and AIStor Free is a free licence with redistribution prohibited rather than open source. The company also sells MemKV, a separate context memory store for AI inference, priced against inference infrastructure.

Founded 2014 · Redwood City, California · min.io