The situation

Hardware prices are climbing and don’t expect relief anytime soon. Buy your technology sooner rather than later. No easing is expected until at least 2027.

Driven by flash media and DRAM price increases, pricing will be a major shock to anyone procuring almost any computerized electronic or storage device for some time. This is an entirely new market, very similar to the GPU supply crunches we have seen for some years now.

SAVE POINT

███████████████████████████████
█  WARNING: PRICE PATCH NOTES  █
█  - "WAIT TO BUY" NERFED      █
█  - "BUDGETS" TAKING DAMAGE   █
█  - "LEAD TIMES" SPAWNING     █
███████████████████████████████

Old strategies still work sometimes. Just not in this dungeon.

This will be hard for many to adjust to, as storage and RAM prices generally trend downward and conventional wisdom would indicate that postponing purchases would make things less expensive if measured by costs over capacity.

In the past two calendar quarters we have seen DRAM procurement costs rise as much as 172% and SSD media raw materials (NAND wafers) are up 246% year over year. DRAM pricing for compute platforms (routers/switches, storage devices, compute platforms, GPUs) accounts for 30–50 percent of costs on these items.

VDURA has indicated that large SSD (32TB) drive prices have increased by 257% (from $3,062 to $10,950) in the past two quarters. Current cost increases are lagging market and are being cushioned by current inventories, meaning that customers have not seen the worst of this yet.

Why are prices increasing?

Enterprise SSD demand is surging, driven by AI data center needs. NAND foundry allocations to SSD storage have exceeded smartphone supply demands for the first time. Manufacturers of DRAM and NAND are prioritizing their AI data center buyers. These buyers want more capacity than the manufacturers can provide.

Manufacturers are NOT expanding capacity, resulting in a classic ‘demand exceeding supply’ scenario. These price increases are structural. While the current spike is unusually steep, I don’t expect the market to return to previous levels or trends. The curve should flatten somewhat once this initial shock passes.

Global buyers are buying components that are not manufactured yet, for devices not yet assembled, for data centers that are not built yet.

COIN ROOM

┌──────────────────────────────┐
│  SUPPLY vs DEMAND METER      │
├──────────────────────────────┤
│  SUPPLY : ████▒▒▒▒▒▒▒▒▒▒     │
│  DEMAND : ████████████████   │
│  STATUS : "STRUCTURAL BOOST" │
└──────────────────────────────┘

When demand stays buffed and supply stays capped, prices don’t “reset” politely.

What can you expect in the near future?

• QLC NAND is expected to increase 60–100% more this calendar year.
• TLC media will increase more slowly, with the main effects seen mid-year.
• Spillover effects will increase HDD prices by as much as 40–50% this year.
• An Account Executive for a Fortune 500 technology company said they anticipate DRAM pricing to double every month for the foreseeable future.
• Another AE indicated active quotes are being pulled — new quotes are being issued with only 14-day windows before requotes are needed.

MUSHROOM POWER

███████████████████████████████████
█  WELCOME TO THE STORAGE SHOP!    █
█  TODAY'S SPECIAL:                █
█   - QLC ↑↑ (60–100%)             █
█   - HDD ↑   (40–50%)             █
█   - QUOTES: 14 DAYS ONLY         █
█  TIP: "COLD DATA != GOLD DATA"   █
███████████████████████████████████

Procurement tip: assume volatility; shorten decision loops where you can.

LET’S TALK ABOUT WHAT THIS MEANS FOR STORAGE – FIND EFFICIENCIES

The DRAM, SSD and HDD increase in costs mean that customers are going to pay MUCH MUCH more for storage this calendar year, with SSD media increasing more than traditional spinning disk. The flash-storage-only companies promise of SSD becoming as cheap as HDD will NOT come to fruition anytime soon.

There seem to be two main choices for storage buyers right now.

1. Don’t buy storage. With the typical organization increasing storage use by 20 percent per year, combined with ‘just in time’ procurement this is not an option for many unless they want to destroy existing data or have not yet turned on data reduction. I am uncertain that any current pricing guarantees/long term contracts will be honored under these market conditions.
2. Find efficiencies. This will mean data reduction and/or tiering data to less expensive media while preserving flash media only for data that needs the performance of SSD.

ITEM GET!

┌───────────────────────────┐
│  TOOLKIT UNLOCKED!        │
├───────────────────────────┤
│  [1] Compression          │
│  [2] Deduplication        │
│  [3] Tiering Policies     │
│  BONUS: "Data Awareness"  │
└───────────────────────────┘

The fastest win in a supply crunch is usually “do more with what you already own.”

Data reduction

Data reduction can be categorized into two methods. Each can be done inline as data is written or post process after data is committed to disk. Inline data reduction often comes with performance penalties on writing data. The efficiencies found from these technologies are very dependent on the type of data you have.

Tiering

Tiering is the ability to place data on different media. The best way to do this is with a policy of some kind, referencing file type, workload, or last accessed time. An example is a solution with SSD and HDD storage that keeps the latest data on SSD and moves data to HDD as it ages.

While there are many third-party products that can move data around and provide links or stubs to the original location of the data, organic solutions native to storage seem to work best. I would recommend a customer-created policy driven solution, as each organization’s needs would be different.

SOME COMMON UNSTRUCTURED STORAGE SOLUTIONS (CURRENT SUPPLY CRUNCH)

Dell PowerScale is one of the most flexible solutions in the market. Dell offers policy-driven compression and deduplication both inline and post process. Dell offers extensive tiering support both within the storage cluster and off cluster to object storage. Dell is one of the few companies that will write guarantees on data reduction rates. Dell offers QLC, TLC and HDD media in PowerScale clusters.

NetApp ONTAP platforms also feature inline/post process compression, inline/post process deduplication and tiering. ONTAP still relies on RAID technologies, which are not as efficient as their competitor’s erasure coding solutions. NetApp has tiering support as well, but policy support is lacking compared to Dell PowerScale and they only tier to S3 object storage. NetApp offers a variety of media for ONTAP clusters with limitations. NetApp also offers a storage guarantee, but less robust than Dell’s.

Pure FlashBlade//S only supports proprietary QLC media with no tiering as we know it. They do offer ‘Zero-Move Tiering’, but this seems to be a throttling solution, allowing performance classes of storage by allocating less compute and networking to entire file systems that are deemed ‘cooler’. Pure FB//S only offers inline compression with no policy support on tiering.

VAST Data is hugely reliant upon SCM and QLC media – both will be impacted by the current supply situation. While VAST Data claims the best data reduction, using compression and deduplication, it is inflexible and inline only. VAST often recommends turning data reduction off in high-write workloads. VAST has no tiering support whatsoever to different media, and in fact their business model is ‘anti-tiering’.

Qumulo does not have data reduction, rather claiming the same efficiencies for the most part that other erasure code-based systems provide. Qumulo does very limited tiering using hybrid nodes with SSD and HDD drives with automated (no policy support) hot/cold block placement. Qumulo can tier between differing clusters in the same distributed namespace, but it does not seem to be able to tier between HDD and SSD nodes in the same cluster.

HOW DO THESE RANK AND WHY?

1. Dell PowerScale Flexible tiering operations, can mix media in a cluster, compression and deduplication with a guarantee, and a supply chain advantage.
2. NetApp ONTAP Similar features, but RAID inefficiencies hurt them. They cannot tier within a cluster, and tiering policy support is limited.
3. Qumulo No data reduction, but can tier between clusters of differing media types. Limited supply chain control compared to larger incumbents.
4. VAST Data Strong data reduction helps, but no native tiering and heavy reliance on constrained media makes competition harder under new pricing realities.
5. Pure FlashBlade//S Compression-focused SSD-only clusters with limited automated tiering; proprietary media dynamics make timing and magnitude of spikes uncertain.

CLEAR STAGE!

┌───────────────────────────┐
│   LEVEL UP! PROCUREMENT    │
├───────────────────────────┤
│  + Tiering Policies        │
│  + Reduction Guarantees    │
│  + Faster Decision Cycles  │
│  - "Wait & See" Strategy   │
└───────────────────────────┘

If the market is the boss, you want buffs—not vibes.

SHARPEN YOUR AX!

It’s going to be a wild ride in 2026 and 2027 for storage purchasers. If your data supports data reduction – you need to be evaluating possible efficiencies NOW, maybe attempt to ride this out if you can.

If you are invested in a storage vendor that does not do tiering, you may be looking at a third-party data manager to move your cooler data to a less-expensive storage solution. If your data is not reduceable or if your workload requires all-flash storage performance, you are at a disadvantage in storage costs.

© TreeBeard Consulting • ONST Technologies