Just How Much Have SSD and RAM Prices Actually Risen?

Our last post talked about rising storage prices in general terms: AI infrastructure spending, Gartner forecasts, TrendForce contract percentages. Useful context, but none of it tells you what a real SSD actually cost last year versus what it costs now. So we went and found that instead, using the same tracked products over time rather than industry-wide averages, and the shape of the story turned out to be more interesting than “prices went up.”

Two very different stages

The first stage runs from 2023 into 2024, and it’s mostly a story about NAND flash recovering from oversupply, not AI. A tracked panel of popular SSD models on Amazon US shows the pattern clearly: a 2TB Samsung 990 Pro averaged around $300 in January 2023, fell to about $150 by that July, and stayed roughly there through the end of 2023, before climbing back to around $180 by August 2024 as manufacturers cut production to restore profitability.

That trough matters for how you read every “prices doubled!” headline from 2025 onward: a lot of that increase is measured from an unusually cheap starting point, not from a stable baseline. It’s a real increase, but the size of the percentage depends heavily on when you start counting.

The second stage is where it gets genuinely severe

The second stage is late 2025 into early 2026, and this is the part that’s harder to explain away with “it was cheap before.” Using fixed, same-product cohorts tracked by GamersNexus, here’s what actually happened between November 2025 and March 2026:

Four months. Not a full year, four months, and 2TB NVMe SSDs went up 112% on average across four tracked models, while a DDR4 memory cohort rose 118%. HDDs rose too, but far less severely, up about 26% for a 16TB drive over a longer window. That gap between HDD and everything else is itself informative: it lines up with NAND flash and DRAM being the components most directly competed for by AI hardware, while hard drives run on a more separate manufacturing and supply chain.

AI is a real driver, but it’s not the only one

It’s tempting to point at the AI buildout and call it a day, and there’s genuine evidence behind that instinct. Micron said its entire calendar-2025 HBM output was sold out with pricing already fixed, months before 2025 even started, and by early 2025 it was already negotiating 2026 supply agreements. SK hynix said something similar about its 2024 HBM output almost a year earlier, in February 2024. OpenAI announced a Stargate-related partnership with Samsung and SK hynix aiming for up to 900,000 DRAM wafer starts a month, a scale normally associated with entire national industries, not a single company’s supply chain. Western Digital described firm purchase orders from its seven biggest customers running through all of 2026, with agreements stretching into 2027 and 2028.

That’s a real pattern: large buyers locking in supply years ahead, which leaves less flexible capacity for everyone else. But it’s not the whole explanation. DDR4 memory specifically saw some of the sharpest percentage increases in the entire dataset, and DDR4 isn’t the memory type AI accelerators actually use, HBM and DDR5 are. DDR4’s price surge is better explained by manufacturers deliberately winding down production of an older product and reallocating that capacity, which is a supply decision, not an AI-demand one. NAND flash also spent 2023 in genuine oversupply, so part of the 2024 recovery was suppliers correcting a mistake, not responding to new demand. And consumer SSD contract prices actually fell during early 2025 on high inventories, before rising again, which is a reminder that this hasn’t been one smooth upward line even in the recent stretch.

The honest summary: AI-driven capacity competition is a real and significant factor in what SSDs and RAM cost right now, but it’s layered on top of an ordinary supply-cycle correction, a legacy-product phase-out, and normal retail volatility. Treating all of it as “the AI boom did this” overstates one true cause at the expense of several other true ones.

What this changes about the advice

Not much, honestly, and that’s sort of the point. Whether the reasons behind higher prices are 30% AI and 70% supply cycle, or the other way around, the practical answer for someone with an archive to protect is the same one we wrote about last time: figure out which of your files actually need a second full copy, verify the copies you already have instead of assuming they’re fine, and use something like PAR2 parity data to add real redundancy to files that don’t strictly need a full duplicate. See our earlier post on that for the actual workflow, this one was mainly about making sure the numbers behind it hold up to scrutiny.

They do. The prices went up. It’s just more complicated, and more interesting, than a single headline can carry.