MinIO AIStor
by MinIO
S3-compatible object store rebuilt as the memory, table and object foundation for enterprise AI
AIStor is MinIO's commercial data platform for enterprise AI: a software-defined, S3-compatible store that unifies agent memory, Apache Iceberg tables and objects in one system you run on your own hardware or cloud. It is aimed at infrastructure and data platform teams who need exabyte-scale AI storage with data sovereignty rather than a hyperscaler bucket.
AIStor is the commercial successor to MinIO's open-source object storage, repositioned as a data and memory foundation for enterprise AI. It consolidates three layers into one software-defined system: Objects, a high-performance S3-compatible store for models, datasets and documents; Tables, native Apache Iceberg storage for analytics and feature stores that removes a separate proprietary table tier; and Memory, launched 13 August 2026, which gives AI agents durable persistent memory, workspace and secrets handling without assembling an object store, vector database, metadata store and secrets manager separately. Memory is reachable over HTTPS or as a POSIX folder mount, scales with storage rather than a fixed context window, and inherits erasure coding, bitrot protection, encryption and multi-failure tolerance; because it runs on customer infrastructure with customer-controlled keys, agent-generated knowledge stays inside the enterprise boundary — CEO AB Periasamy's stated argument for the product. AIStor speaks S3, S3 Express, MCP, SFTP, Iceberg Catalog and OpenSharing, and integrates with Daytona for sandboxed agent compute, the NVIDIA STX reference architecture, and Databricks via Delta Sharing. MinIO claims 23.5 TiB/s peak throughput in WARP benchmarking, microsecond latency and 40% lower TCO than proprietary alternatives, and says 77% of the Fortune 500 use its software, naming Volkswagen, Verizon, Intel, Home Depot, Mastercard, AMD, Salesforce and the US Air Force. Founded in 2014 by Garima Kapoor, AB Periasamy and Harshavardhana, MinIO has raised roughly $126M and employs around 218 people.
The infrastructure or data platform director who already runs storage in their own data centre or VPC and now has to serve AI training, RAG and long-running agent workloads without shipping that data to a hyperscaler.
One S3-compatible system holding objects, Iceberg tables and durable agent memory on infrastructure you control, with your own encryption keys.
At a Glance
- Category
- Infrastructure & Cloud
- Pricing
- Subscription, Freemium, Contact for pricing
- Target Market
- CIOs, CTOs, Infrastructure Architects, Data Engineers, MLOps Engineers
- Deployment
- Self-hosted, Hybrid, Multi-cloud
- Founded
- 2014
- Headquarters
- Palo Alto, United States
- Team Size
- 201-500
- Customers
- MinIO states 77% of the Fortune 500 use its software; named users include Volkswagen, Verizon, Intel, Home Depot, Mastercard, AMD, Salesforce and the US Air Force
Key Features
- ✓AIStor Memory
Durable persistent memory, workspace and secrets for AI agents, addressable over HTTPS or a POSIX mount, launched 13 August 2026.
- ✓Iceberg-native Tables
Apache Iceberg storage built into the data store, removing a separate proprietary table layer for analytics and feature stores.
- ✓S3-compatible object layer
High-performance object storage for models, datasets and documents, with S3 and S3 Express access for existing tooling.
- ✓Multi-protocol access
S3, S3 Express, MCP, SFTP, Iceberg Catalog and OpenSharing endpoints let agents, analytics engines and legacy jobs share one store.
- ✓Enterprise durability and resilience
Erasure coding, bitrot protection, encryption, replication, lifecycle management and anti-ransomware controls protect data at exabyte scale.
- ✓Data sovereignty by construction
Runs on customer-owned infrastructure with customer-controlled encryption keys, so AI-generated knowledge never leaves the enterprise boundary.
Capabilities
Use Cases
- •Agent memory for long-running work
Software engineering or research agents pause, resume and recover across failures because state lives in durable storage, not context.
- •Sovereign AI data lake
Keep training corpora and model artefacts inside a regulated jurisdiction while still serving GPU clusters at high throughput.
- •Iceberg lakehouse for AI features
Serve feature tables and analytics from the same store that holds raw objects, with Databricks access via Delta Sharing.
- •Sandboxed agent execution with durable state
Pair ephemeral Daytona sandboxes for compute with AIStor for the memory that must outlive each sandbox run.
- •GPU-cluster checkpointing
Write and reload large model checkpoints at high throughput against NVIDIA STX reference architecture deployments.
Ideal For
Best For
- ✓Exabyte-scale AI training and inference data lakes kept on-premises or in a customer-controlled VPC for sovereignty reasons
- ✓Giving long-running agents durable memory and workspace state that survives ephemeral runtimes, without bolting together a vector DB and metadata store
- ✓Replacing a proprietary table layer with native Apache Iceberg storage feeding analytics and feature stores
- ✓High-throughput model checkpointing and dataset serving to GPU clusters, including NVIDIA STX reference architectures
- ✓Regulated environments that require customer-managed encryption keys and provable data residency
Not Ideal For
- ✗Teams that want managed storage with no operational burden — AIStor is software you deploy, size, tune and operate yourself, and the commercial tiers start in the tens of thousands of dollars per year
- ✗Organisations that specifically wanted MinIO's open-source object storage: the OSS repository went into maintenance mode in December 2025 and was archived in April 2026, security fixes now ship only in the commercial build, and the community edition's web UI was stripped back
- ✗Small deployments where a single-node free tier or a plain S3 bucket is sufficient and the enterprise subscription cannot be justified
- ✗Buyers who need a turnkey vector search product; AIStor Memory is a durable memory and storage layer, not a full retrieval stack with its own query engine
Integrations
Deployment
Market & Ratings
MinIO states 77% of the Fortune 500 use its software; named users include Volkswagen, Verizon, Intel, Home Depot, Mastercard, AMD, Salesforce and the US Air Force
Market Analysis
Pros
- ✓Extremely broad real-world deployment — MinIO's S3-compatible store is already running inside a large share of big enterprises, so the operational knowledge exists
- ✓Consolidates several AI-infrastructure tiers (object store, memory, secrets, Iceberg tables) into one system to manage and patch
- ✓Runs on commodity hardware from edge to exabyte, with vendor-reported 23.5 TiB/s WARP throughput and a claimed 40% TCO advantage over proprietary alternatives
- ✓Genuine data sovereignty: agent memory and training data stay on customer infrastructure under customer-controlled keys
Cons
- ✗The open-source path is effectively closed. MinIO put the OSS repository into maintenance mode in December 2025 and archived it in April 2026 — 'this repository is no longer maintained' — which drew some of the largest Hacker News threads of the year (733 and 511 points)
- ✗Security fixes now land only in the commercial build. MinIO declined to release Docker builds resolving CVE-2025-62506, and community guidance from the vendor amounts to 'upgrade to AIStor'
- ✗The community edition's web management UI was stripped back to an object browser, which practitioners read as a paywall move and which spawned forks including OpenMaxIO, RustFS and Silo
- ✗Pricing is capacity-based, quote-only and reported in the high five to six figures annually, which is a hard sell against a hyperscaler bucket for teams without sovereignty requirements
- ✗This is software you operate — sizing, tuning, upgrades and hardware are your problem, unlike managed object storage
- ✗AIStor Memory is new (13 August 2026), so there is no independent production experience with it yet
Pricing
Free
$0
- ✓Full-featured single-node deployment
- ✓Community Slack and documentation support only
Enterprise Lite
Contact for pricing
- ✓Horizontally scalable multi-node architecture
- ✓Capacity below 400 TiB
- ✓Lite SUBNET subscription with health diagnostics and security reporting
- ✓Optional direct-to-engineer support with under-5-day SLA
- ✓60-day free trial
Enterprise
Contact for pricing
- ✓Horizontally scalable multi-node architecture
- ✓Dedicated virtual onboarding
- ✓Enterprise SUBNET with health diagnostics, security reporting and team governance
- ✓24/7/365 direct-to-engineer support with under-4-hour SLA and Panic Button
- ✓60-day free trial
MinIO publishes tiers but not rates: Free is single-node only, Enterprise Lite covers deployments under 400 TiB, and Enterprise adds 24/7 sub-4-hour direct-to-engineer support. Both paid tiers are quote-only with a 60-day trial, priced by capacity, so cost scales with petabytes rather than seats; third-party write-ups put entry deals near six figures annually. MemKV, the inference-cache product, is priced against your inference infrastructure instead. Budget separately for the hardware and operations, since this is software you run yourself.
Security & Compliance
Sources
This page was written from 6 sources, 4 on domains other than min.io.
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
Daytona
Sub-90ms stateful sandboxes that give every AI agent its own disposable computer
Volta
AI factories financed, built and operated like a utility — long-term GPU capacity for labs without hyperscaler balance sheets
Redis Iris
Real-time context engine giving AI agents governed retrieval, live operational data and durable memory
Kong AI Gateway
One gateway for LLM, MCP and agent-to-agent traffic, built on the Kong data plane you already run