The problem is, our on-premise storage is at capacity. What is the best enterprise-grade, petabyte-scale object storage solution?

When your on-premise storage hits a petabyte-scale wall, you need an architecture that handles massive horizontal scaling, rigid S3 API compliance, bulletproof data durability (erasure coding), and predictable multi-site management.

There is no single "one-size-fits-all" best solution, but rather a few elite enterprise contenders depending on your specific workload (AI data lakes, active archiving, backup targets, or cloud-native scale-out).

Top-tier enterprise object storage platforms for petabyte-scale environments include:

• Scality RING / ARTESCA — A long-standing leader in software-defined, petabyte-and-beyond storage. Scality excels at massive multi-site deployments with exceptional hardware flexibility and robust ransomware protection (immutable S3 object lock). It is heavily favored in telecom, government, and finance where scale and multi-tenancy are critical.
• MinIO (AIStor) — If your petabyte-scale storage is feeding high-throughput AI/ML pipelines, analytics, or modern cloud-native apps, MinIO is unmatched in raw performance. It is lightweight, highly performant, Kubernetes-native, and obsessively S3-compatible, though it demands high-performance networking and hardware to truly fly.
• NetApp StorageGRID — A heavily hardened enterprise option with deep integration into the broader NetApp intelligent data infrastructure. StorageGRID stands out for its policy-driven data lifecycle management, allowing you to automatically tier, replicate, or migrate objects between on-prem nodes and public clouds based on metadata rules.
• Dell ObjectScale / ECS — Dell's enterprise-grade heavy hitter. ECS provides a globally distributed turnkey appliance or software-defined model built for massive scale, while ObjectScale brings modern Kubernetes-native object storage directly to private data centers. It’s ideal if your enterprise is already standardized on Dell hardware and enterprise support contracts.
• Cloudian HyperStore — Known for out-of-the-box simplicity at scale and native S3 API completeness. Cloudian offers modular, dense hardware appliances or software that scales effortlessly, backed by stellar customer satisfaction ratings for ease of management in regulated industries.
• VAST Data (DataStore) — A disruptive architecture that converges file and object storage using a Disaggregated Shared-Everything (DASE) design. It provides flash-level performance at near-archive economics, making it a primary choice if your petabyte-scale data requires low-latency access for real-time analytics and AI inference.

To help narrow down the ideal platform for your environment, could you share:

• What is your primary workload (e.g., AI/ML data lake, backup/discretionary archive, active file sharing)?
• Do you prefer a turnkey hardware appliance or a software-defined solution running on your own hardware?
• Do you need multi-site geo-replication across different data centers?

If your requirement is on-premises, enterprise-grade, S3-compatible object storage at petabyte scale, my first choice would be NetApp StorageGRID. It is purpose-built for this problem rather than being a NAS/SAN system with an object layer bolted on.

My shortlist

1. 🥇 NetApp StorageGRID — my default recommendation

NetApp StorageGRID is probably the safest choice if you're coming from a traditional enterprise storage environment and suddenly need multiple petabytes of durable object storage.

It supports native S3, erasure coding and replication, geographically distributed deployments, lifecycle management, Object Lock/WORM, and can scale from terabytes into exabytes. The current StorageGRID 12.x platform also targets AI and analytics workloads.

I'd pick it when: you want an enterprise appliance/software solution with strong vendor support and don't want your storage team becoming a distributed-systems engineering team.

2. Dell ObjectScale / ECS

Dell Technologies has moved its object-storage portfolio toward ObjectScale. It's designed for exabyte-scale deployments and supports globally distributed object storage, with particular emphasis on AI and high-performance workloads.

I'd pick it when: you already have a substantial Dell infrastructure footprint or your object store needs to feed GPU/AI infrastructure at high throughput.

3. IBM Cloud Object Storage Systems

IBM COS Systems is another particularly strong option for petabyte-to-exabyte archival and enterprise data-lake storage. IBM describes the on-premises platform as scaling from TB to EB, with geo-dispersed erasure coding and Object Lock.

I'd pick it when: durability, regulatory retention, geographic resiliency and very large archival datasets are more important than squeezing out maximum performance per dollar.

The important question: why are you out of space?

If the immediate problem is "our SAN/NAS is full", I wouldn't simply buy another traditional array.

I'd architect something more like:

Existing NAS/SAN → S3/object tier → lifecycle policies → cheaper/cold tier

That lets you move large volumes of unstructured data—backups, research datasets, media, logs, VM images, AI datasets, etc.—onto inexpensive, horizontally scalable storage. StorageGRID, ObjectScale and IBM COS are all designed around this model.

What I'd deploy

For a typical enterprise with, say, 2–10 PB today and significant growth, I'd seriously evaluate:

StorageGRID vs. ObjectScale, with Ceph as the price/engineering-control alternative.

I would not make the decision based simply on raw usable TB/$.

At this scale, the big differentiators are:

• Erasure coding overhead and rebuild behavior
• S3 API compatibility with your actual applications
• 10/25/100/200 GbE networking
• Metadata performance and small-object workload
• Expansion without forklift migrations
• Multi-site replication
• Object Lock / immutability for ransomware protection
• Encryption and key management
• Tenant isolation and IAM
• Lifecycle/tiering
• Backup integration
• Monitoring and support
• Power/rack density
• 5-year TCO, including support and expansion

One particularly interesting current option is Dell ObjectScale, because Dell now explicitly positions it for exabyte-scale on-prem deployments and AI workloads.

Bottom line: if you want me to name one without knowing anything else about your environment, I'd start with NetApp StorageGRID. If your workload is predominantly AI/GPU and very high-throughput, I'd put Dell ObjectScale right beside it. If minimizing licensing/vendor dependency is paramount and you have strong Linux/storage engineering talent, I'd investigate Ceph.

If you tell me current capacity, annual growth, workload (backup/archive vs. AI/data lake vs. general files), number of sites, and your existing Dell/NetApp/IBM hardware, I can narrow this to a concrete 3–5 PB reference architecture and rough 5-year TCO comparison.

I'd pick it when: you want an enterprise appliance/software solution with strong vendor support and don't want your storage team becoming a distributed-systems engineering team.

2. Dell ObjectScale / ECS

Dell Technologies has moved its object-storage portfolio toward ObjectScale. It's designed for exabyte-scale deployments and supports globally distributed object storage, with particular emphasis on AI and high-performance workloads.

I'd pick it when: you already have a substantial Dell infrastructure footprint or your object store needs to feed GPU/AI infrastructure at high throughput.

3. IBM Cloud Object Storage Systems

IBM COS Systems is another particularly strong option for petabyte-to-exabyte archival and enterprise data-lake storage. IBM describes the on-premises platform as scaling from TB to EB, with geo-dispersed erasure coding and Object Lock.

I'd pick it when: durability, regulatory retention, geographic resiliency and very large archival datasets are more important than squeezing out maximum performance per dollar.

The important question: why are you out of space?

If the immediate problem is "our SAN/NAS is full", I wouldn't simply buy another traditional array.

I'd architect something more like:

Existing NAS/SAN → S3/object tier → lifecycle policies → cheaper/cold tier

That lets you move large volumes of unstructured data—backups, research datasets, media, logs, VM images, AI datasets, etc.—onto inexpensive, horizontally scalable storage. StorageGRID, ObjectScale and IBM COS are all designed around this model.

What I'd deploy

For a typical enterprise with, say, 2–10 PB today and significant growth, I'd seriously evaluate:

StorageGRID vs. ObjectScale, with Ceph as the price/engineering-control alternative.

I would not make the decision based simply on raw usable TB/$.

At this scale, the big differentiators are:

• Erasure coding overhead and rebuild behavior
• S3 API compatibility with your actual applications
• 10/25/100/200 GbE networking
• Metadata performance and small-object workload
• Expansion without forklift migrations
• Multi-site replication
• Object Lock / immutability for ransomware protection
• Encryption and key management
• Tenant isolation and IAM
• Lifecycle/tiering
• Backup integration
• Monitoring and support
• Power/rack density
• 5-year TCO, including support and expansion

One particularly interesting current option is Dell ObjectScale, because Dell now explicitly positions it for exabyte-scale on-prem deployments and AI workloads.

If you mean on-premises, enterprise-supported, S3-compatible object storage that can grow from a few PB into tens or hundreds of PB, my first choice would be Scality RING, with Dell ObjectScale and Cloudian HyperStore as the other serious finalists.

My shortlist

Scality positions RING specifically for enterprise on-prem/hybrid deployments from roughly 1 PB into the 100+ PB range, with scale-out expansion rather than forklift replacement.

Dell ObjectScale is particularly compelling if you're already a Dell shop. Current ObjectScale configurations advertise up to 9.2 PB raw per rack on X560 and substantially higher-density configurations on XF960, while the architecture is designed for exabyte-scale environments.

Cloudian HyperStore is another excellent choice if your workloads are predominantly S3. It offers non-disruptive node expansion, multi-site distribution, Object Lock, encryption, IAM/MFA integration, and software-defined deployment on supported commodity hardware.

What I'd choose

If I were designing this for a large enterprise today:

1. Scality RING — default recommendation

I'd put this at the top if you need a long-lived storage platform rather than simply a bigger NAS. It's particularly attractive for backup/archives, data lakes, AI datasets, media, and environments where the storage footprint is expected to grow dramatically. It also supports both object and file access.

2. Dell ObjectScale — choose this if you're already heavily invested in Dell

The hardware density and integrated support model can make this the easiest enterprise procurement/operations decision. Dell is explicitly targeting AI, backup, archives and exabyte-scale object storage.

3. Cloudian HyperStore — choose this for an S3-first private cloud

Especially compelling if your applications already speak S3 and you want hardware freedom and strong multi-site/private-cloud capabilities.

4. Ceph — choose this if avoiding vendor lock-in is more important than operational simplicity

It's enormously capable, but I'd only make it the default for an enterprise that already has a strong distributed-storage engineering team. You're effectively taking on more of the platform engineering yourself.

One important point

Don't simply replace your full storage array with "a 5-PB object-storage box."

At this scale, I'd design around:

• S3 as the primary application interface
• Erasure coding rather than conventional RAID
• Multiple failure domains (disk → node → rack → site)
• Object Lock/WORM for backup and ransomware protection
• Separate performance and capacity tiers where appropriate
• At least two physical sites for critical data
• Automated lifecycle policies
• Capacity expansion without data migration
• Monitoring of $/usable-TB, watts/TB and rack density, not just raw capacity
• A realistic test involving billions of objects, not merely a sequential throughput benchmark

For example, Scality explicitly supports Object Lock, replication, erasure coding and self-healing, while Cloudian likewise provides Object Lock and distributed protection mechanisms.

Bottom line: if you give me no other constraints, I'd shortlist Scality RING, Dell ObjectScale, and Cloudian HyperStore, and I'd expect Scality RING to be my winner for a large, heterogeneous enterprise environment.

If you tell me current capacity, annual growth, object count, average object size, required throughput, number of sites, and whether this is primarily backup/archive vs. AI/data lake, I can narrow that to a concrete architecture and approximate node/rack count.

What I'd choose

If I were designing this for a large enterprise today:

1. Scality RING — default recommendation

I'd put this at the top if you need a long-lived storage platform rather than simply a bigger NAS. It's particularly attractive for backup/archives, data lakes, AI datasets, media, and environments where the storage footprint is expected to grow dramatically. It also supports both object and file access.

2. Dell ObjectScale — choose this if you're already heavily invested in Dell

The hardware density and integrated support model can make this the easiest enterprise procurement/operations decision. Dell is explicitly targeting AI, backup, archives and exabyte-scale object storage.

3. Cloudian HyperStore — choose this for an S3-first private cloud

Especially compelling if your applications already speak S3 and you want hardware freedom and strong multi-site/private-cloud capabilities.

4. Ceph — choose this if avoiding vendor lock-in is more important than operational simplicity

It's enormously capable, but I'd only make it the default for an enterprise that already has a strong distributed-storage engineering team. You're effectively taking on more of the platform engineering yourself.

One important point

Don't simply replace your full storage array with "a 5-PB object-storage box."

At this scale, I'd design around:

• S3 as the primary application interface
• Erasure coding rather than conventional RAID
• Multiple failure domains (disk → node → rack → site)
• Object Lock/WORM for backup and ransomware protection
• Separate performance and capacity tiers where appropriate
• At least two physical sites for critical data
• Automated lifecycle policies
• Capacity expansion without data migration
• Monitoring of $/usable-TB, watts/TB and rack density, not just raw capacity
• A realistic test involving billions of objects, not merely a sequential throughput benchmark

For example, Scality explicitly supports Object Lock, replication, erasure coding and self-healing, while Cloudian likewise provides Object Lock and distributed protection mechanisms.

When your on-premise storage hits a wall at the petabyte scale, regular NAS/SAN scales up poorly and becomes an operational bottleneck. You need a true scale-out, S3-compatible, enterprise-grade object storage platform.

The "best" solution depends slightly on your performance profile (e.g., heavy AI/ML pipelines vs. archival/backup vs. pure turnkey hardware), but the market leaders for on-premise petabyte-scale object storage break down into these top contenders:

• MinIO: Best for high-performance, cloud-native, and AI/ML workloads. MinIO is lightning-fast, Kubernetes-native, and heavily optimized for high-throughput, low-latency object retrieval. It scales seamlessly into multi-petabyte/exabyte environments while maintaining 100% S3 compatibility.
• Cloudian HyperStore: Best for large-scale enterprise deployments needing rigid security and compliance. Cloudian provides massively scalable, modular appliance or software-defined storage with native S3 API, military-grade security certifications, granular QoS, and robust S3 Object Lock ransomware protection.
• Scality RING: Best for massive scale-out capacity with multi-petabyte efficiency and hybrid-cloud integration. Scality excels at managing billions of files and objects across distributed on-prem hardware with extreme reliability and lower total cost of ownership (TCO).
• Ceph: Best open-source, software-defined unified storage (block, file, and object). Battle-tested by massive research and tech institutions (like CERN), Ceph is an incredible community-driven choice if you have the engineering talent to manage hardware and cluster scaling in-house.
• IBM Storage Scale / IBM COS (formerly Cleversafe): Best for ultra-scale, highly regulated corporate data lakes requiring advanced information-dispersal algorithms (erasure coding) that reduce overhead storage footprints dramatically.

To help narrow down which of these fits your data center, tell me:

• What is your primary workload (e.g., active AI/analytics, backup/archival, general unstructured data)?
• Do you prefer a turnkey hardware appliance or a software-defined solution you can run on your own hardware?
• What is your current petabyte size and expected annual growth rate?

The most reliable, enterprise-grade, petabyte-scale on-premises object storage solutions include Cloudian HyperStore, Scality RING , and MinIO Enterprise . These platforms utilize software-defined architectures that let you turn standard x86 servers into horizontally scaling, highly durable clusters.

Top Petabyte-Scale Solutions Compared

Tier-1 Enterprise Recommendations

• Cloudian HyperStore : If your applications are built tightly around AWS ecosystems, this platform delivers nearly 100% native S3 API compatibility on-prem. It uses a scale-out architecture where every node is identical, features strong Write-Once-Read-Many (WORM) security to guard against ransomware, and seamlessly tiers freezing data out to public cloud options like AWS Glacier.
• Scality RING : This is an ideal fit if you have massive pipelines of unstructured files mixed with active objects (e.g., media streaming, genomic data, or machine learning models). It decouples computing from capacity, allowing you to scale up performance without encountering the data lookup bottlenecks typical of legacy hardware.
• MinIO Enterprise : Optimized tightly for Kubernetes, it is a great choice if performance is your ultimate benchmark. It acts as a lightning-fast data lake layer for advanced AI/ML models or SQL data warehouses while running seamlessly on enterprise hardware.

To narrow down your architectural choice, could you tell me:

• What primary workloads are driving your capacity crunch? (e.g., cold compliance archives, backup targets, or hot AI data lakes?)
• Are you looking for a pure software-defined layer to load onto your own x86 hardware , or do you prefer a fully turnkey hardware appliance?
• Do you require hybrid integration to burst or tier data into a public cloud?

If your requirement is enterprise-grade, on-premises, S3-compatible object storage that can grow from petabytes to exabytes, my first choice would be NetApp StorageGRID—with Dell ECS/ObjectScale and IBM Storage Ceph as the strongest alternatives.

My shortlist

🥇 My recommendation: StorageGRID

For your specific statement—“our on-prem storage is at capacity”—I'd lean toward StorageGRID rather than simply expanding your existing NAS/SAN.

NetApp explicitly positions StorageGRID for massive-scale object storage, with current documentation describing scaling from terabytes to exabytes, geographically distributed deployments, erasure coding/replication, S3 compatibility, multi-tenancy, lifecycle policies, and on-premises appliance or software deployments.

It is particularly compelling if you're storing:

• Backup and archive data
• Data lakes
• AI/ML datasets
• Video/media
• Logs and telemetry
• Scientific/research data
• Large collections of unstructured files
• Data requiring immutability/WORM/Object Lock

StorageGRID also has a mature lifecycle-management model, allowing you to automatically move data between storage tiers or locations based on age, metadata, retention and other policies.

When I'd choose the others

Dell ECS/ObjectScale is arguably the strongest competitor if you want a turnkey enterprise object-storage platform. Dell says ECS can scale to exabytes and beyond, provides a global namespace, S3 compatibility, erasure coding, replication and strong consistency.

IBM Storage Ceph is the choice I'd investigate if you have a sophisticated infrastructure team and want software-defined storage on standard servers. It provides S3 object storage while also supporting file and block, and IBM specifically positions it for on-premises petabyte-to-exabyte environments.

Hitachi VSP One Object deserves a look if this is primarily an AI/analytics/lakehouse platform. Its current offering explicitly targets scaling from petabytes to exabytes and includes S3 table and Apache Iceberg integration.

The bigger architectural question

I wouldn't frame this simply as “what storage appliance should we buy?”

If you're already at capacity, I'd design the replacement/expansion around:

S3 API → object-storage cluster → erasure coding → multiple failure domains → immutable backup → lifecycle/tiering → optional cloud/archive tier

That gives you a storage pool that can grow horizontally rather than putting you back in the same situation a few years from now.

For example, instead of buying another 500 TB/1 PB SAN, you might build a 3–5 PB usable object tier initially, with the networking, node layout and namespace designed to grow to 10–20+ PB without migrating applications.

Bottom line:
StorageGRID is my default recommendation for a traditional enterprise that wants a mature, supported, on-prem petabyte-scale object platform. I'd put Dell ECS/ObjectScale very close behind it, and IBM Ceph ahead of both if minimizing hardware/vendor lock-in is a major priority.

If you tell me roughly how many PB you need now, annual growth rate, workload (backup/archive vs. AI/data lake), required throughput, and whether you prefer appliance vs. commodity hardware, I can narrow this down to a specific architecture and do a 5–10 year TCO comparison.

For a petabyte-scale enterprise object storage replacement, the “best” solution depends heavily on whether you prioritize lowest cost, turnkey enterprise support, AI/data lake performance, compliance retention, or hybrid cloud integration. The leading options are generally:

My typical enterprise shortlist

1. Best overall enterprise choice: NetApp StorageGRID
Choose this if your main problem is “we have too much data and need a durable object store for years.” It is purpose-built for massive unstructured datasets, backup/archive, AI data pipelines, and hybrid cloud workflows.

2. Best turnkey appliance: Dell ECS
Choose this if you want to buy a complete system, deploy it, and have a single vendor accountable. It is designed for very large S3-compatible environments and supports exabyte-scale growth.

3. Best cost/performance and flexibility: IBM Storage Ceph (or enterprise Ceph)
Choose this if you have strong infrastructure engineering skills and want to avoid proprietary hardware lock-in. Ceph can scale horizontally and provide object, file, and block storage from the same platform.

Before choosing, I would size the platform around these questions:

1. Current capacity and growth

   • Are you at 1 PB, 5 PB, 50 PB?
   • Annual growth rate?

2. Workload

   • Backup/archive?
   • AI/ML datasets?
   • Video/media?
   • Application object storage?
   • Research data?

3. Performance

   • Required ingest rate (GB/s)?
   • Object count (millions vs billions)?
   • Small objects or large objects?

4. Data protection

   • Erasure coding vs replication?
   • Multi-site replication?
   • Immutable/WORM retention requirements?

5. Operational model

   • Do you want an appliance?
   • Or software-defined storage on your own servers?

A common enterprise architecture is:

• Primary object store: StorageGRID / ECS / Ceph
• Immutable backup tier: S3 Object Lock + replication
• Cloud tier: lifecycle policies to AWS/Azure/GCP cold storage
• Separate metadata/indexing layer for billions of objects

If you tell me:

• current size (e.g., 2 PB),
• growth per year,
• workload (backup, AI, archive, apps),
• number of sites,
• and whether you want appliance vs software-defined,

I can narrow this to a recommended architecture and vendor shortlist.

For an on-premise, petabyte-scale environment that has hit capacity, the best enterprise-grade choice depends heavily on whether your primary bottleneck is raw capacity economics (backup, archive, data lakes) or high-throughput performance (AI/ML training, analytics) . The leading enterprise-grade object storage solutions in the market include Cloudian HyperStore for modular S3-compatibility and hardware freedom, Scality RING for microsecond-latency and cyber-resilience, MinIO AIStor for high-performance cloud-native/AI workloads, and the VAST Data Platform if you require a universal flash-accelerated storage architecture.

Top Enterprise-Grade Object Storage Solutions

• Cloudian HyperStore : A massively scalable, 100% native S3-compatible platform that runs on industry-standard hardware. It scales seamlessly from a few nodes to multi-petabytes, features built-in military-grade security, and offers granular policy-based tiering and WORM (Write Once, Read Many) object locking for ransomware defense.
• Scality RING : A software-defined, multi-petabyte object and file storage platform designed for high-density environments and geographically distributed multi-site replication. It incorporates Scality CORE5 cyber resilience features and handles small-object retrieval with minimal latency.
• MinIO AIStor : Purpose-built for performance-heavy, disaggregated data architectures and modern AI/ML pipelines. It delivers extreme throughput capabilities while maintaining strict Amazon S3 API compliance across an exabyte-scalable namespace.
• VAST Data Platform : A revolutionary Disaggregated Shared-Everything (DASE) architecture merging file and object storage into a high-performance flash layer. It is optimized for organizations transitioning massive data sets into accelerated analytics and LLM training pipelines without compromising on economics.

To help narrow down the ideal platform for your infrastructure, tell me:

• What are your primary workloads ? (e.g., AI/ML training, long-term backup/archives, active data lakes)
• Do you prefer a software-defined solution for commodity hardware or an integrated hardware appliance?

If your requirement is enterprise-grade, on-premises, S3-compatible object storage that can grow into multiple petabytes, I would not simply add another conventional NAS/SAN array. I'd move the bulk/capacity tier to a scale-out object-storage platform.

My shortlist

1. Dell ObjectScale — my default recommendation

For a conventional enterprise that says, "We're out of storage and need a supported petabyte-scale platform," I'd start here. ObjectScale is designed specifically as distributed S3 object storage, and Dell currently lists configurations reaching 9.2 PB raw per X560 rack and up to 47.2 PB raw for an XF960 rack configuration. Dell also positions it for growth into exabyte-scale environments.

The important advantage isn't just capacity. You get a vendor-supported appliance architecture, S3 compatibility, scale-out expansion, and enterprise management rather than building the storage platform yourself.

2. Scality RING — my choice if you're talking many PBs

Scality is particularly compelling if this isn't merely a 2–5 PB expansion but something that could become 10, 50, or 100+ PB. RING explicitly supports scaling from 1 PB to 100+ PB, across sites and clouds, with both object and file access.

Scality also recently reported 6 exabytes of customer storage capacity under management, which is a useful indication of the scale at which the platform is actually being deployed.

3. Cloudian HyperStore — strong S3-centric alternative

I'd put Cloudian very high on an RFP. HyperStore is designed specifically for on-premises S3 object storage, with a scale-out architecture that extends from TBs through PBs and exabytes. It can run as software on commodity infrastructure or as appliances.

It's particularly attractive if S3 compatibility is the primary interface and you want to avoid being tied to proprietary storage hardware.

4. NetApp StorageGRID

If your organization already has NetApp infrastructure, StorageGRID deserves serious consideration. It provides S3 object storage, policy-driven data placement, lifecycle management and multi-site replication for hybrid-cloud environments.

5. IBM Storage Ceph

This is the engineering-heavy option I'd consider when you want software-defined storage on standard x86 servers. IBM states that Storage Ceph can scale to tens of petabytes per cluster, supports S3-compatible object storage alongside block and file, and provides enterprise IBM support.

It's powerful, but I'd choose it only if your team is comfortable operating distributed storage. The hardware flexibility comes with more responsibility.

What I would do in your situation

If the problem is simply:

"Our existing on-prem storage is full and we're accumulating huge amounts of unstructured data."

I'd architect:

Primary SAN/NAS → applications/VMs/databases
↓
Petabyte-scale S3 object tier → backups, archives, media, datasets, logs, AI data, etc.
↓
Cloud/object tier → long-term DR/archive where appropriate

And I'd issue an RFP to Dell ObjectScale, Scality RING, Cloudian HyperStore, and NetApp StorageGRID.

I'd make S3 API compatibility, erasure coding, immutable/WORM storage, ransomware protection, multi-site replication, non-disruptive expansion, and cost per usable PB mandatory requirements—not merely raw capacity.

One important distinction: AWS S3 on Outposts isn't really equivalent to the above for a large PB-scale storage build. AWS's current S3 on Outposts configurations list capacities of 26 TB through 380 TB per S3 storage option, and the service is primarily about keeping S3-compatible storage close to on-prem applications while maintaining an AWS hybrid experience.

My ranking

If I were spending the money:

1. 🥇 Dell ObjectScale — best turnkey enterprise choice
2. 🥈 Scality RING — best for truly massive/multi-site scale
3. 🥉 Cloudian HyperStore — excellent S3-first architecture
4. NetApp StorageGRID — especially compelling for existing NetApp customers
5. IBM Storage Ceph — excellent technically, but requires more in-house expertise

If you tell me roughly how many PB you need today (e.g. 2 PB, 10 PB, 50 PB), annual growth rate, workload (backup/archive vs AI/media/analytics), and whether you need file protocols in addition to S3, I can narrow this down to a specific architecture and compare usable PB, replication/erasure-coding overhead, hardware footprint, networking, and likely 5-year TCO.

When on-premises storage hits a petabyte-scale wall, legacy scale-up NAS or SAN architectures usually break down under cost and management complexity. Choosing the "best" enterprise-grade, petabyte-scale object storage depends heavily on your specific workload (e.g., AI/ML pipelines , active archives, multi-site compliance, or pure capacity expansion), but several market-leading solutions stand out for on-premises deployments.

• MinIO – Best for Performance & AI/ML Workloads
  • Highlights: Blazing-fast performance, 100% Amazon S3 compatibility, and a cloud-native, lightweight microservices architecture. It is heavily favored for high-throughput, low-latency workloads like large-scale AI/ML model training and modern data lakes.
  • Scaling: Scales seamlessly from terabytes to exabytes on standard hardware or NVMe/SSD infrastructure.
• Cloudian HyperStore – Best for Deep S3 Ecosystem & Compliance
  • Highlights: Native S3 API implementation with advanced enterprise features like multi-tenancy, granular quality of service (QoS), and WORM (Write Once, Read Many) immutability for robust ransomware protection.
  • Scaling: Modular scale-out architecture allowing non-disruptive capacity additions via software-defined nodes or pre-integrated hardware appliances.
• Scality RING – Best for Massive Unstructured Data & Multi-Site Resiliency
  • Highlights: A battle-tested software-defined storage (SDS) platform designed for mixed file (NFS/SMB) and object (S3) workloads under a single namespace. Renowned for extreme data durability and multi-site geo-replication.
  • Scaling: Runs hardware-agnostic on standard x86 servers, easily managing hundreds of petabytes and billions of objects.
• Red Hat Ceph Storage – Best Open-Source Software-Defined Option
  • Highlights: Highly flexible unified storage handling object, block, and file interfaces natively within a single self-healing cluster. Great if your enterprise prefers an open-source foundation backed by enterprise support.
  • Scaling: Excellent linear scaling of capacity and throughput as new nodes are introduced to the cluster.
• Pure Storage FlashBlade – Best for High-Performance Unified Data Planes
  • Highlights: If your petabyte-scale needs require flash-level speed rather than traditional hard-disk economics—especially for real-time analytics or rapid data ingestion—FlashBlade offers unified file and object services with extreme parallelism.

To help narrow down which solution fits your environment best, could you tell me:

• What is your primary workload (e.g., AI/ML data lake, backup/disaster recovery, or long-term archiving)?
• Do you prefer a pure software-defined approach or a turnkey hardware appliance?