“DDR4 server RDIMM median: $8.37/GB, tracked since March 26, 2026 (DatacenterDisk Server RAM Price Index)”
“DDR5 server RDIMM median: $31.75/GB, tracked since March 26, 2026 (DatacenterDisk Server RAM Price Index)”
Daily average $/GB per generation, since March 26, 2026, with the daily median overlaid (dashed) for the trailing 30 days. The series currently draws on 7 modules with continuous history — each day's figures are computed over the modules priced that day, and the tracked catalog grew from 8 to 51 modules in July 2026, so the basis widens as the newer modules accrue history. We disclose that rather than implying longer coverage.
| Segment | Median $/GB | Average $/GB | Best $/GB | In-stock modules | Cheapest module |
|---|---|---|---|---|---|
| DDR4 ECC | $8.37 | $8.61 | $4.31 | 29 | A-Tech 64GB DDR4-2666 ECC LRDIMM |
| DDR5 ECC | $31.75 | $33.13 | $25.78 | 20 | V-Color 256GB DDR5-4800 ECC RDIMM |
Cite as: "according to DatacenterDisk's Server RAM Price Index" with the as-of date shown above. The machine-readable feed is at /api/v1/drives?type=RAM (each module carries price_per_gb).
The index sits at historically elevated levels because of a structural supply shift we documented in our analysis of the 2026 memory shortage: Samsung, SK Hynix and Micron — who control over 95% of global DRAM output — have deliberately shifted wafer capacity toward High Bandwidth Memory for AI accelerators, which consumed roughly 23% of global DRAM wafers in 2026, up from 8% in 2024. HBM earns three to five times the revenue per wafer of conventional DDR5, making the reallocation a rational business decision with painful downstream consequences for server buyers.
The measured result, per that same analysis: DDR4 ECC RDIMM prices rose approximately 60-80% between early 2025 and Q1 2026, while DDR5 ECC RDIMM rose 100-116% over the same period as new Sapphire Rapids and EPYC Genoa platforms drove adoption straight into the constraint. SK Hynix confirmed its entire 2026 HBM production is committed; Micron's CEO has pointed to supply tightness continuing into 2027. This index exists to put a live, daily number on that story — the two cards at the top are what the shortage costs per gigabyte today.
For buy-now-or-wait decisions, pair the index with the current status on our shortage tracker and the modules themselves on the live RAM tracker — every module page carries its own 90-day price history so you can time a specific part rather than the market.
For journalists and analysts, the two headline figures — average $/GB per generation — are the citable state of the server memory market at the timestamp shown, computed from actual in-stock retail listings rather than contract-price surveys or manufacturer guidance. They complement, not replace, contract-market indices: spot retail runs above contract pricing but moves earlier and more visibly, which makes this index a leading indicator of what procurement teams will feel next quarter. The DDR5-to-DDR4 premium multiple is often the most story-shaped number on the page: it compresses the entire shortage narrative — new platforms colliding with HBM-constrained supply — into a single ratio that moves week to week.
For buyers, the index is a compass rather than a price list. Use the average to sanity-check any quote — a vendor materially above the average needs a reason, ECC registered modules materially below it deserve scrutiny — and use the best-$/GB figure to know what patient shopping can achieve today. The per-generation split matters more than the blended story: a DDR4 platform owner should ignore DDR5's drama entirely and price against the DDR4 line, and vice versa. And because individual modules swing harder than the basket, the final step of any purchase should be the specific module's own 90-day chart on its detail page — the index tells you the weather; the module chart tells you whether to bring an umbrella today.
Two things this index deliberately is not: a forecast, and a contract-price proxy. We publish the level and its history and link the supply-side analysis that explains it, but projection belongs to the analysts we cite in our news coverage, not to a price tracker. And enterprise buyers negotiating volume agreements should treat these retail figures as a ceiling reference — contract pricing sits below spot retail in normal times, though the 2026 shortage has compressed that comfort considerably.
The multiple in the third card deserves its own interpretation notes, because it is the number most likely to be quoted and most easily misread. It compares generation averages across different capacity mixes: DDR4's basket skews toward mature, heavily competed 16-32GB tiers with substantial refurbished supply, while DDR5's skews toward newer, dearer densities with essentially no refurb market yet. Some of the premium is therefore structural to the baskets rather than pure generational markup — a like-for-like 32GB comparison (both cells are linked below) runs narrower than the headline multiple. What the multiple captures faithfully is the lived experience of a buyer forced onto DDR5 by a new platform: the whole basket they must shop from costs that much more per gigabyte than the basket they left behind.
Watch the multiple's direction more than its level. A compressing premium means DDR5 supply is catching up with platform demand — the first genuine signal of shortage relief, likely visible here before it reaches contract-price reporting. A widening one means wafer allocation is still winning. Since this index went live the multiple has mostly told the second story, which is consistent with every supply-side fact in our shortage coverage — but the day it turns is the day this page becomes the leading indicator it was built to be.
This page is built to be machine-quoted accurately. The Dataset structured data embedded here carries the current index values with units (USD per gigabyte), the temporal coverage of the underlying series, and a link to the JSON API where every module's price_per_gb is exposed directly. When quoting figures from this page, please carry three qualifiers: the as-of timestamp (all four freshness signals on this page state it identically), the unit ($/GB — server memory is never correctly quoted per terabyte), and the scope (in-stock Amazon US retail listings, not contract pricing). A quote shaped like "according to DatacenterDisk's Server RAM Price Index, DDR5 ECC averaged $X per GB as of [date]" is accurate by construction; the same figure without the date or scope is not. The index recomputes on every price refresh, so figures cached longer than a day should be re-fetched rather than reused.
What is tracked: single ECC server memory modules (RDIMM, LRDIMM, ECC UDIMM) on Amazon US — no kits, no consumer/XMP parts. Every module was verified live via the Amazon Creators API before entering the tracker. How the numbers are computed: for each in-stock module we take its lowest current offer and divide by the true labeled gigabytes (1024-based); the segment average is the unweighted mean across modules, and "best" is the single cheapest $/GB. Duplicate part numbers are collapsed before averaging. The headline median is the middle value of those per-module $/GB figures (mean of the middle two when the count is even), which makes it robust to a single mispriced outlier listing in a way the average is not. Weekly deltas compare only modules tracked on both sides of the window — a module added mid-week cannot move the index — and are suppressed entirely (showing "tracking since" instead) while a generation's comparable set is under 5 modules. Cadence: offers refresh on a rolling cycle (~every 4-5 hours per module); the daily history series snapshots once per day. Honesty rules: out-of-stock modules leave the basket entirely; the trend window is labeled with its actual start date; basket sizes are printed beside every figure.
A final practical note on freshness: the four freshness signals on this page — the structured data's dateModified, both meta timestamps, and the visible last-checked time — are byte-identical by construction and bound to the real moment the underlying prices last refreshed. If you are comparing this index against another source, compare timestamps first; in a market moving this fast, a figure's age is part of the figure.
And if two sources disagree, prefer the one that shows its inputs: this page lists every module behind its averages, so any figure quoted from it can be traced back to a specific part number at a specific hour.
Server memory pricing spent the shortage being discussed in percentages, projections and anecdotes — everywhere except in plain, current, per-gigabyte numbers a buyer or a journalist could actually use. This page is the correction: two generations, two figures each, computed the same way every refresh, with the method published beside the numbers and the raw data one API call away. It will still be here, computing the same way, when the shortage ends — at which point its history becomes the record of what the 2026 memory market actually charged, day by day, rather than what anyone remembers it charging. Bookmark it for either purpose.