Every HPE Array Line Before B10000 Is Discontinued. Support Runs to 2030.

If you run HPE 3PAR, Nimble, Primera, Alletra 9000 or Alletra 6000, the array you own can no longer be ordered new. That is not a warning about something coming — the last-order dates have passed. What you have instead is a long, orderly runway: support on the Alletra generation runs to the end of 2030 on the published schedule. That combination, no new purchases but years of support, is exactly the shape of a problem that gets deferred until it becomes urgent, and it is why storage refresh budget gets approved late and spent badly.

This page explains what happened to each line, where it lands, what the successor actually is architecturally — it is not simply a newer array — and the order in which to make the decision. We sell storage, so the practical bias here is toward getting you a specified configuration rather than a discovery call.

Four families, one successor

Two separate product lineages — the mission-critical block line descended from 3PAR, and the midrange line descended from Nimble — converge on one platform. HPE's own refresh material states it plainly: "With your HPE 3PAR, HPE Nimble Storage, or HPE Primera all-flash storage array at end-of-life…".

Alongside it sits a second, different machine: the Alletra Storage MP X10000, an object and file platform for AI, analytics, active data lakes and fast backup. It is not a replacement for a block array and should not appear in a like-for-like refresh quote. More on where it does belong below.

The dates

Published lifecycle milestones for the Alletra 6000 and 9000 lines, which are consistent across lifecycle trackers:

The Alletra 5000 runs a year behind, with published end-of-life on 1 May 2026 and formal obsolescence on 31 October 2026.

Verify before you budget. These dates come from third-party lifecycle trackers, not from a document we can read at source — HPE publishes them in Product Change Notifications on a support portal that renders through JavaScript and cannot be quoted directly. The authoritative documents are PCN a00149116en_us (Alletra 9000) and a00159357en_us (Alletra 5000), plus the end-of-availability notice dp00005509en_us covering Alletra 6000, Alletra 5000 and Nimble. Ask HPE or us to send you the current PCN for your exact model before a date goes into a budget paper. This is the same discipline we apply to reading vendor end-of-support data generally: the lookup method is more durable than a republished table.

Two practical readings of that timetable. First, 31 May 2026 already passed — new hypervisor and operating-system releases are no longer being validated against these arrays, so a platform upgrade on the host side can quietly move you outside the support matrix well before 2030. Second, the two upgrade end-of-sale dates matter more than the final one for anyone whose capacity is still growing: after 2028 you cannot buy your way out of a full array.

What the B10000 actually is

The important thing about the successor is that it is not simply a faster array of the same shape. It is a disaggregated platform, and that changes how you size and how you buy.

• Controllers and capacity scale independently. In switched configurations the system supports 2, 3 or 4 controller nodes, and controllers no longer have to be deployed as matched HA pairs. You can add performance without adding capacity, and capacity without adding controllers.
• Capacity. Up to 24 NVMe SSDs per shelf, in TLC from 1.92 TB to 30.72 TB and QLC from 15.36 TB to 30.72 TB. HPE quotes maximum raw capacity of 5.9 PB per switched system and maximum effective capacity of 16.6 PB at an assumed 4:1 compaction ratio.
• Entry point. A minimum of eight TLC SSDs per system, with drive upgrades in two-drive increments. The floor is lower than most people expect from a platform positioned against mission-critical arrays.
• Connectivity. Fibre Channel, NVMe-oF/FC, NVMe-oF/TCP and iSCSI. NVMe over TCP is the one to note: it delivers NVMe semantics over standard Ethernet, which changes the fabric conversation for anyone who was going to buy Fibre Channel by default.
• Personas. Block and file on the same platform, with object handled by the X10000 line.

Practically, this means a refresh is no longer a like-for-like swap of one array model for its successor model. It is a sizing exercise with two independent axes, and the right answer is often a smaller controller count than the array you are replacing plus more capacity behind it — or the reverse. A supplier quoting you "the B10000 equivalent of your 6000" has skipped the exercise.

The guarantees, read carefully

HPE attaches two commercial guarantees to the platform: a 100% data availability guarantee and a 4:1 total data savings guarantee. Both are real contractual commitments rather than marketing adjectives, and both come with conditions — on configuration, on supported software versions, and on what counts as a qualifying outage or a qualifying dataset.

The 4:1 figure in particular does work in the capacity arithmetic above: the 16.6 PB effective capacity number assumes it. If your data is already compressed, encrypted at the application layer, or dominated by media, you will not see 4:1, and your effective capacity will be closer to raw. Size on your own measured reduction ratio if you have one, and get the guarantee terms in writing before you let a compaction assumption shrink the capacity you order. The broader vendor claims that circulate alongside — up to 40% lower cost, 98% operational time savings — are vendor-published figures against a composite baseline, and belong in a business case only with that baseline named.

Migration: what is actually supported

HPE documents array-to-array migration paths rather than expecting a host-side copy, and the tooling differs by source:

• From 3PAR — Peer Motion, HPE's federation feature for online, non-disruptive migration of hosts, host sets, virtual volumes and volume sets. It requires a Peer Motion licence and management through SSMC, and it is unidirectional.
• From Primera or Alletra 9000 — a documented migration path driven from the HPE GreenLake for Block Storage UI.
• From Nimble, Alletra 5000 or 6000 — a different lineage and a different data layout, so confirm the supported method for your specific source array and software version rather than assuming the 3PAR path applies.

Three questions decide whether the migration is a weekend or a quarter, and none of them are answered by the array datasheet: how many hosts and volumes are in scope, whether the source array can run alongside the target long enough to move data online, and whether your Fibre Channel fabric has the ports and zoning headroom for both arrays at once. Get those confirmed before you accept a migration estimate from anyone, us included.

Where the X10000 fits — and where it does not

The X10000 is object and file storage speaking S3 and S3a, built on the same disaggregated NVMe architecture, aimed at AI and machine learning pipelines, active data lakes, high-performance computing, real-time analytics and fast backup and restore. Its distinguishing component is a data intelligence node that extracts metadata and vector embeddings inline as data lands.

It is the right answer for a data lake, a backup target at scale, or a retrieval pipeline feeding a model. It is the wrong answer for the block workloads on your existing array — virtual machines, databases, anything presented as a LUN. If a quote replaces a 3PAR or Nimble with an X10000, the supplier has misread the requirement. Where both genuinely apply, they are two line items, not one.

The decision, in order

1. Establish your real dates. Get the current PCN for your exact model in writing. Everything downstream is scheduling.
2. Measure before you size. Current usable capacity, actual growth rate over the last 24 months, real data reduction ratio, peak IOPS and latency requirement. Sizing a disaggregated platform from the old array's model number wastes the main advantage it has.
3. Decide the fabric. Fibre Channel again, or NVMe over TCP on Ethernet. This has more effect on the total project cost than the array line item, and it is easiest to decide before the array is quoted.
4. Separate block from object. B10000 for the block estate. X10000 only if you have an actual object or analytics requirement, priced separately.
5. Sequence against the upgrade end-of-sale dates, not the final support date. If capacity is growing, the 2028 hardware-upgrade cut-off is your real deadline.
6. Decide what happens to the old array. Decommissioned enterprise storage retains value, and that value is a credit against the refresh — see the secondary market and residual value.

Frequently asked questions

Can I still buy HPE Nimble or Alletra 6000?

No. Discontinuance for the Alletra 6000 and 9000 lines was 30 November 2025 on the published schedule, and Nimble was superseded before that. Anything offered to you as new stock now is either remaining inventory, secondary-market hardware, or mislabelled — worth establishing which, using the checks in sourcing HPE.

How long is my current array supported?

On the published schedule, engineering support and the support period for the Alletra 6000 and 9000 lines end 31 December 2030, with hardware upgrades ending 31 December 2028 and software upgrades 31 December 2029. Older 3PAR and Primera systems are on earlier schedules that vary by model. Confirm your exact model against HPE's Product Change Notification before treating any of these as your date.

Is the B10000 a drop-in replacement for my array?

Functionally it covers the same block workloads, but it is architecturally different: controllers and capacity scale independently, and controller counts of 2, 3 or 4 are supported without matched HA pairs. That makes a like-for-like model swap the wrong way to size it. Expect a genuine sizing exercise from measured data, not a conversion table.

What is the difference between the B10000 and the X10000?

B10000 is block and file for mission-critical and general enterprise workloads. X10000 is object and file — S3 and S3a — for AI, analytics, data lakes and fast backup, with a data intelligence node that generates metadata and vector embeddings inline. They share an architecture and address different requirements. A block array refresh is a B10000 conversation.

Do I have to move to Fibre Channel?

No. The B10000 supports Fibre Channel, NVMe-oF/FC, NVMe-oF/TCP and iSCSI. NVMe over TCP runs NVMe on standard Ethernet, which is worth pricing against an FC refresh if your fabric is also ageing — the switches and HBAs are frequently a larger line than people expect when they budget only for the array.

Is the migration non-disruptive?

HPE's Peer Motion is documented as online and non-disruptive for the 3PAR path, and Primera and Alletra 9000 migrate through the GreenLake for Block Storage UI. Whether your migration is non-disruptive depends on host operating systems and versions, multipathing configuration, fabric capacity to run both arrays concurrently, and application constraints. Treat "non-disruptive" as a property of the tool, and validate it against your estate before committing to a cutover window.

Should we consider a different vendor at refresh?

It is the natural moment to ask, since the migration effort is comparable whichever direction you go. The honest counterweight is that staying within HPE preserves the documented migration paths and your team's operational knowledge, and moving vendors does not. If you want the comparison run properly, NetApp versus Pure Storage covers the main alternatives and we will quote across vendors on the same requirement.

Related

• How to choose HPE ProLiant servers and storage — platform selection across the HPE portfolio
• HPE support tiers explained — why four-hour response is weaker than six-hour repair
• HPE iLO and Compute Ops Management licensing — the remote console is a per-server licence
• HPE memory population rules — why 12 DIMMs per processor runs at half the bandwidth of 8
• Sourcing HPE · part numbers decoded · GreenLake vs owning
• Enterprise storage supply · NetApp vs Pure Storage · Lenovo storage vs NetApp
• Reading vendor end-of-support data — the method, and why we do not republish dates as fact
• Secondary market and residual value — what the outgoing array is worth against the refresh
• HPE ProLiant Gen11 vs Gen12 · DL360 vs DL380 Gen12 — if servers refresh alongside the array
• Buying at project scale · stock vs BTO/CTO
• HPE stock, lead times and pricing — current availability and firm quotes

Sources

• HPE — Alletra Storage MP B10000 QuickSpecs ("2-Node and 4-Node storage controllers", switched configurations of "2, 3, or 4"; 24 NVMe SSDs per shelf; TLC 1.92–30.72 TB and QLC 15.36–30.72 TB; minimum "eight (8) TLC SSDs per system", two-drive increments; 5.9 PB raw and 16.6 PB effective per switched system at "4:1 data compaction ratio"; Fibre Channel, NVMe-oF/FC, NVMe-oF/TCP and iSCSI; "100% Data Availability Guarantee")
• HPE — Refresh storage technology to Alletra Storage MP B10000 (3PAR, Nimble and Primera named as the arrays at end of life; "4:1 total data savings guarantee"; independent scaling of performance and capacity)
• HPE — Alletra Storage MP X10000 (object and file, S3 and S3a; AI, machine learning, active data lakes, backup and restore, HPC and real-time analytics; data intelligence node generating metadata and vector embeddings inline)
• HPE — Peer Motion (online, non-disruptive migration of hosts, host sets, virtual volumes and volume sets; requires a Peer Motion licence and SSMC; unidirectional)
• Lifecycle milestone dates as published by third-party lifecycle trackers. Authoritative source: HPE Product Change Notifications a00149116en_us (Alletra 9000), a00159357en_us (Alletra 5000) and end-of-availability notice dp00005509en_us (Alletra 6000, Alletra 5000, Nimble Storage) — request the current version for your model before budgeting against a date.