Run the math on a petabyte-scale ransomware recovery and the backup repository — not the backup software, not the network — is almost always the choke point. A dedupe disk appliance that happily ingests 40TB an hour can crawl at a fraction of that when it has to rehydrate and stream data back out, and at petabyte scale the difference is measured in days of downtime. That is the decision moment this brief is about: QLC flash has pulled all-flash backup targets down to roughly disk-level acquisition cost, and that changes who should still be buying spinning disk for their repository tier.
Below: what changed in the QLC economics, why restore throughput is now the ransomware SLA that matters, honest numbers on where dedupe disk still wins, and vendor-by-vendor reads on Pure Storage, NetApp, and Veeam.
What changed
For a decade the objection to flash as a backup target was one word: price. QLC NAND ended that argument at the acquisition line. Pure Storage’s FlashBlade//E — its capacity-optimized, QLC-based file and object platform — is positioned explicitly as an unstructured data repository priced comparably to disk-based systems, with Pure claiming operating costs around 60% lower than disk once power, space, and admin time are counted. NetApp answered from the other direction: its AFF C-Series (C250, C400, C800) puts cost-optimized QLC flash under the full ONTAP stack and aims it squarely at secondary storage, backup targets, and capacity workloads, with NetApp citing petabyte-range capacities per cluster.
Meanwhile the density curve keeps working in flash’s favor — Pure has publicly targeted 300TB DirectFlash modules on its 2026 roadmap, densities hard-drive vendors cannot answer. The verdict: as of mid-2026, “flash costs too much for backup” is no longer automatically true. It is a workload-by-workload calculation, and that is new.
Why it matters: restore throughput is the new ransomware SLA
Backup windows stopped being the constraint years ago — incremental-forever chains and changed-block tracking saw to that. The constraint that boards now ask about is the clean-room recovery: how fast can you restore everything, from a known-good copy, after an encryption event. That is a full-fidelity, sequential-read-at-scale problem, and it is exactly where dedupe disk architectures are weakest, because every restored block has to be rehydrated from deduplicated segments scattered across spindles.
Flash targets invert that. Pure has claimed restore rates up to 270TB per hour on large FlashBlade configurations — vendor-published and scale-dependent, so treat it as a ceiling, not a promise. But even discounting heavily, the gap against a rehydrating disk appliance is not 20%; it is often an order of magnitude on mass restores. If your recovery SLA math prices a day of downtime in the millions — and for most enterprises above a few thousand employees it does — the repository’s read path is the SLA. Our 2026 RTO/RPO benchmark data makes the same point from the other end: the enterprises missing recovery targets are overwhelmingly bottlenecked on restore reads, not on backup software.
The economics, honestly
Here is the honest counterweight: purpose-built dedupe appliances still win raw $/TB at rest, and it is not close when dedupe ratios run high. Backup data dedupes exceptionally well, and an appliance built around inline dedupe can land effective cost per terabyte well below any all-flash system. If your repository exists to satisfy retention policy and restores are rare, small, and non-urgent, dedupe disk remains the rational buy.
| Dedupe disk appliance | QLC all-flash target | Generic disk repository | |
|---|---|---|---|
| $/TB at rest | Best in class with high dedupe ratios | Roughly disk-comparable to acquire; vendors claim lower TCO on power/space | Cheap to buy, costly to power and rack at scale |
| Mass restore throughput | Weakest — rehydration penalty grows with scale | Strongest — sequential reads at flash speed | Middling; spindle-bound |
| Instant-recovery / live-mount performance | Generally poor | Production-adjacent | Usable for a handful of VMs |
| Best fit | Long retention, infrequent restores | Aggressive recovery SLAs, ransomware clean-room plans | Small estates, tight capital budgets |
The takeaway a VP can repeat: dedupe disk optimizes the cost of keeping data; flash optimizes the cost of getting it back. Price your downtime and the spreadsheet picks the tier for you. For the fuller cost model, see our backup storage cost-per-TB analysis for 2026.
Vendor read: Pure Storage
Pure made the aggressive bet here. FlashBlade//E is a purpose-built play for exactly this tier — scale-out file and object, QLC DirectFlash modules, and a pricing posture designed to kill the disk comparison at the quote stage. Strengths: genuine restore performance at scale, the Evergreen subscription model that takes forklift refreshes off the table, and operational simplicity that shows up in real admin-hours savings. Where it is weaker: Pure’s economics depend on its density roadmap holding, the entry configuration is sized for petabyte-class estates rather than mid-market ones, and buyers should pressure-test the claimed TCO deltas against their own power and facilities costs rather than accepting vendor math. Pure fits the enterprise that has decided recovery speed is a board-level requirement and wants a dedicated repository platform to deliver it.
Vendor read: NetApp
NetApp’s C-Series is the incumbent-friendly path. You get QLC economics under ONTAP, which means the tooling, SnapMirror replication, and SnapLock immutability your team already runs — a real advantage when the backup target must slot into an existing compliance and replication fabric. Strengths: one operating environment across primary and secondary tiers, mature immutability, and strong effective-capacity claims once ONTAP’s dedupe and compression are applied. Where it is weaker: ONTAP is a primary-storage OS carried into a capacity tier, so you pay some complexity tax a purpose-built repository avoids, and C-Series is rarely the cheapest option in a bake-off at rest. NetApp fits the ONTAP shop that wants flash-speed restores without introducing a new vendor or a new operational model. Our Pure Storage vs. NetApp head-to-head goes deeper on where each platform earns its premium.
Where Veeam fits
Veeam does not sell the target, which is precisely why its read matters — the software layer is where flash repositories either pay off or get wasted. Veeam Data Platform’s Scale-out Backup Repository lets you tier a flash performance extent in front of cheaper capacity, so you do not have to buy flash for every retention point — only for the restore-critical window, typically the most recent 14 to 30 days. Its instant-recovery features are the other half of the argument: live-mounting dozens of VMs directly from the repository is a flash-class workload, and it is where disk targets visibly fall over. Veeam’s ONTAP integration also gives NetApp shops snapshot-orchestrated workflows the generic targets do not get. The caution: Veeam’s hardened Linux repository immutability and its fast-clone efficiencies work on any target, so do not let a storage vendor imply flash is required for safe backups. Flash buys speed, not safety — immutability is a software and configuration decision.
What to do about it
- Do the throughput math first. Divide the terabytes you must restore in a clean-room scenario by the hours your business case allows. If the answer exceeds roughly 10TB/hour sustained, test your current appliance’s real restore rate — most disk-era buyers have never measured it.
- Tier, don’t replace. Put flash under the restore-critical window (14–30 days) and keep dedupe disk or object storage for deep retention. Veeam’s SOBR and equivalent constructs make this a configuration exercise, not a migration.
- Price downtime into the TCO. If one avoided day of outage exceeds the 5-year cost delta between flash and disk tiers, the flash target is the conservative buy, not the extravagant one.
- Make vendors prove restore rates on your data. Ingest benchmarks are marketing; demand a proof-of-concept restore of your actual workload mix before signing.
Frequently asked questions
Is all-flash storage good for backups?
Yes, with a caveat. All-flash targets excel when restore speed matters — ransomware recovery, instant VM recovery, aggressive RTOs. For long-retention data that is rarely restored, dedupe disk or object storage still delivers lower cost per terabyte at rest. Most enterprises should tier: flash for the recent restore window, cheaper media behind it.
Is QLC flash reliable enough for backup workloads?
For this workload profile, yes. QLC’s lower write endurance is a poor match for write-intensive primary workloads, but backup targets are sequential-write, read-heavy-on-demand — close to the ideal QLC pattern. Both Pure and NetApp warranty their QLC platforms for these use cases, and media management in the array masks endurance concerns in practice.
What restore speed do I need for ransomware recovery?
Work backward from your tolerable outage. A 500TB critical estate with a 24-hour recovery target needs roughly 21TB/hour sustained — beyond what most dedupe appliances deliver on rehydrated restores. At petabyte scale with a one-to-two-day target, you are in flash-target territory almost by definition.
Does an all-flash backup target replace immutability?
No. Immutability comes from software and configuration — Veeam hardened repositories, NetApp SnapLock, object lock on S3-compatible targets — not from the media type. Flash changes how fast you recover; immutability determines whether you have anything clean to recover from. You need both, and they are separate line items in the design.
Enterprise Techie publishes vendor-honest analysis like this daily — get the brief by email, free.