SSDs got expensive too, and for the same reason as RAM

We wrote earlier this year about why RAM prices jumped. The short version: AI datacenters are buying every spare bit of memory the three big manufacturers can make, and consumer and business hardware is competing for what’s left. The same thing has happened to storage. NAND flash, the chips inside SSDs, is up sharply, and it’s worth a separate note because the dynamics are a little different and the timeline is, if anything, worse.

What actually moved

Through late 2025 and into 2026, NAND contract prices rose hard quarter over quarter, something like 85 to 90 percent in the first quarter of 2026 alone, with another 70-plus percent forecast for the second. Enterprise SSDs led, since those are the drives going into AI servers, and consumer drives followed. A 1 TB consumer SSD that was around $45 in mid-2025 was closer to $90 by early 2026. Enterprise SSD contract prices were up roughly 40 to 50 percent in Q4 2025 and another 30-plus percent on top of that in Q1 2026.

Phison’s CEO said publicly that NAND prices had more than doubled and that the company’s entire 2026 production was already spoken for. When a major controller maker says that out loud, it’s not a forecast. It’s the current state.

Why storage, specifically

DRAM at least has a clean story: HBM eats fab capacity, less conventional DRAM comes out, prices rise. NAND is a bit more roundabout.

The driver is the same, AI infrastructure, but the mechanism is volume, not exotic packaging. An AI server rack needs a lot more storage than a traditional server: somewhere in the range of ten to fifty times as much, depending on whose number you believe and what the rack is doing. Training datasets, checkpoints, vector stores, model weights, logs. Enterprise SSDs are now around 60 percent of global NAND output by bit. The hyperscalers placed huge orders, the manufacturers (Samsung, SK Hynix, Kioxia, Micron, SanDisk) filled them first, and the consumer channel got whatever was left at whatever price clears it.

There’s also a supply-side echo of the RAM situation: NAND makers had a rough 2023 and cut investment, so there wasn’t much slack in the system going into the AI buildout. New capacity is a multi-year project. Nothing meaningful comes online before late 2027.

Who this hits

Anyone speccing servers or NAS. Storage is a real line item again, the way it was a decade ago. If you’re sizing a file server, a backup target, or a NAS refresh, the per-terabyte number you remember is wrong. Re-quote it.

Laptop and prebuilt buyers. The OEMs are passing this through alongside the RAM increases. Configurations are getting smaller: a machine that shipped with a 1 TB SSD a year ago might ship with 512 GB now at the same price.

Anyone running a homelab or a small on-prem setup. The drives are still available; they just cost what they cost. Budget accordingly.

HDD buyers, oddly. Spinning disks haven’t moved nearly as much, and demand has shifted back toward them for bulk cold storage. If you’ve been all-flash by reflex, a hybrid layout (SSD for hot data, HDD for cold) is suddenly worth the design effort again.

What to do

Most of this is the same advice as for RAM, which is itself the same advice as for any supply crunch:

1. Buy what you need now, not what you might need in two years. Over-provisioning storage “to be safe” was cheap insurance when a terabyte cost $40. It isn’t anymore.
2. Quote it fresh. Any storage number in a budget or proposal written before late 2025 is stale. Add a buffer of 30 to 50 percent on SSD line items for 2026 capex, more for enterprise drives.
3. Use the right tier. Not everything needs NVMe. Backups, archives, media libraries, cold data: HDD is fine, and the price gap has widened in its favour.
4. Don’t panic-buy. Drives are not unobtainable. If a vendor is pushing you to over-order “before it gets worse,” that’s a sales tactic, not a supply update.
5. Reconsider cloud versus on-prem for storage-heavy workloads. The major cloud providers have buying power and longer-term contracts that smooth out spot pricing. If you were on the fence about a big on-prem storage build, the math has shifted toward cloud for the next 18 to 24 months, the same way it has for compute.

When it resolves

Same answer as the DRAM post, which is to say: not soon. The capacity isn’t there, the AI demand isn’t slowing, and the new fabs are 2027 to 2028 events. A reasonable planning assumption is that SSD and NAND prices stay elevated through 2027 and start normalising in 2028. Treat it as a structural change to your hardware budget for the next two years, not a one-time surcharge you can wait out.

If you’re buying machines into this market (laptops, workstations, servers), we wrote a companion piece on how to spend sensibly when memory and storage are the expensive parts.

If you’re sizing storage or infrastructure for a project and want a second set of eyes on the numbers, we’re happy to help.