What we track

MemRadar tracks 235 consumer RAM and SSD products sold in the United States: 119 memory kits and modules, and 116 solid state drives. The catalog covers DDR4 and DDR5 memory in desktop and laptop form factors, and NVMe and SATA drives across the common capacities.

The catalog is not a complete census of the market. It was assembled from the products PC builders actually shop for, which means it over-represents mainstream and enthusiast parts and under-represents server memory, industrial storage, and obscure brands. Products are added over time; none have been removed for reasons other than being wrong.

Tracked days counts distinct days on which a price was recorded for that product, so a median of 200 means a typical product in that segment carries roughly 200 recorded days of history.

Where the prices come from

Price history for Amazon.com listings is licensed from Keepa, a long-running price-tracking service with its own historical archive of Amazon.com listings. That archive is why some products on this site have price history predating MemRadar itself, in a few cases back to 2015. Keepa is the source of every historical chart, every all-time low and high, every average on the site, and every figure in the Price Index.

Newegg prices come from Newegg's product catalog feed, distributed through Rakuten Advertising, which MemRadar receives as an approved publisher. The feed carries current prices and availability, not history, so Newegg appears on this site as a live price comparison and never as a line on a chart or in the Price Index.

No prices on this site are scraped. Both sources are licensed or contractual, which is a deliberate choice: a price you cannot explain the provenance of is a price you should not publish.

How often prices update

Prices for Amazon.com listings are scheduled for collection every four hours, up to six times a day. The service that runs these jobs sometimes delays or skips a scheduled run. Between September 24 and October 6, 2026, we recorded an average of about four successful fetches a day. Each fetch writes one observation per product that has a price, and every product page shows how long ago its price was recorded.

Newegg's catalog feed regenerates once daily, so Newegg prices refresh once a day, with a full reconciliation weekly. Fetching more often would re-read the same file.

Scheduled jobs occasionally run late or fail. When that happens the site keeps showing the last recorded price rather than guessing, and the timestamp on the page tells you how stale it is. The site is built to be visibly out of date rather than invisibly wrong.

How the numbers are calculated

Current price. The most recent recorded price for a product. Where the Amazon.com listing has no offer at all, from Amazon or any other seller, the page says the product is unavailable and shows the last price seen, labelled as such.

All-time low and high. The lowest and highest prices in that product's recorded history, with the month they occurred. These are bounds within our data, not claims about every price the product has ever carried anywhere. Where a product has fewer than five recorded prices in total, both are omitted rather than shown, because an extreme drawn from a handful of sightings is not an all-time anything.

90-day average. The mean of a product's recorded prices over the past 90 days. This is the baseline the price line on each product page compares against, which is why that line describes a price as below, in line with or above its recent average, relative to its own recent history rather than to any absolute standard. Where fewer than ten prices were recorded inside that window the average is omitted rather than computed, and the price line that rests on it is omitted with it.

Price per gigabyte. Current price divided by total capacity. For memory kits, capacity means the kit total: a 2x16GB kit is 32GB. Price per gigabyte is compared against the median for that product's segment, calculated fresh at each rebuild from every tracked product in that segment whose capacity we can parse.

Segment medians. Median rather than mean, for every figure computed across a segment. In a market with occasional gray-market listings and outlier pricing, a mean can be dragged by a handful of rows while a median cannot. The 90-day average above is the one deliberate exception, because it describes a single product against itself rather than a population.

How the price index works

The index has two parts. The monthly record is its year-over-year figure: for each product with at least 20 days of price coverage in both months, we take its median daily price in the month and compare it with its own median daily price in the same month a year earlier, and the published figure is the median of those per-product changes, per segment. Each comparison follows one listing, the same Amazon ASIN, across both months. The live market reading is the short-period part, described further down.

Which price counts. The index covers Amazon.com listings, in US dollars, and Newegg prices are never part of it. Each stored observation is one price for the listing, taken from Keepa's series in a fixed order. From July 22, 2026, each scheduled fetch takes the AMAZON series (Amazon's own offer) when it has a price, otherwise the NEW series (a new-condition marketplace price on the same listing), otherwise the BUY_BOX_SHIPPING series (a Buy Box price including shipping), if that series is populated. Before that date the stored history is Keepa's AMAZON series, with its gaps, or its absence, filled from the NEW series; it holds no Buy Box series. MemRadar historically did not retain which price source supplied each observation, so the two months of one comparison can come from different sources. MemRadar does not use dedicated Lightning Deal, Prime-exclusive, or Warehouse price series and applies no coupon, Subscribe & Save, or rebate adjustment. On September 8, 2026, 165 of 225 products with a current price (73.3%) were using the marketplace-new fallback. This is a point-in-time diagnostic, not the marketplace share of historical Price Index observations.

Why the record will not reconcile with the monthly CSV. The monthly series is the median of each month's readings across whatever products were tracked that month. The record compares each product with itself. They answer different questions and differ by design, and there is a second, mechanical reason they cannot be recomputed from one another: before July 22, 2026 the stored history is a one-time load of Keepa's price history, which records a price only when it changes, so a month with four stored rows is a month whose price moved four times, not a month observed four times. The record therefore works from a carried-forward daily series rather than from raw rows.

The carry rule. A product's price holds from one reading until the next reading, until an observed no-offer event, or for at most 90 days, whichever comes first, and a month starts from the latest earlier in-stock price only if no no-offer event lies between and it is within 90 days. "No offer" is recorded two different ways, and a careful reader recomputing from the history will meet both. Before July 22, 2026, a stored row marked out of stock starts a no-offer interval that ends at the next in-stock row. From July 22, 2026, a fetch writes a row for every product that has a price and no row at all for one that does not, so a product that is absent from every fetch on a UTC day that had at least one fetch is treated as having no offer that day; a product present in any fetch that day is covered; and a day with no fetch at all is not evidence either way and leaves the carry running. A product qualifies for a month with at least 20 covered days, and covered days are calendar days contributing to the carried series, never a count of stored observations.

Exclusions. Three short lists of listings are left out of every segment by rule, and each record stores them: a relisted duplicate (the same product under a second ASIN, which would count one product twice), a multi-unit listing (a two-pack, whose price is not one drive's price) and a drive sold with a service plan. The service-plan exclusion is definitional rather than numerical: the bundle is a different product from the drive alone, and it is excluded whether it sells above or below the plain drive. Hardware variants, such as a heatsink version of a drive, stay in.

Survivorship. MemRadar's catalog was assembled in July 2026, so every historical figure describes products that were still on sale then. Products discontinued before that date are not represented, and a segment's record for an earlier month is a statement about the products that survived to be tracked, not about everything sold in that month.

The freeze and corrections. For months after the launch series, a record is computed once, on the first regeneration after a price fetch timestamped past the end of the month, and is then frozen: it is never recomputed, and MemRadar keeps the paired listings and prices each record was computed from. The twenty-four records from October 2024 to September 2026 are the launch series: they were reconstructed from the stored price history by the same rule, frozen together on October 12, 2026, and are marked as reconstructed. If a published figure is later found to be wrong, the original stays as published and a correction is added as a dated entry with the original and revised values and the reason. From September 2026 onward, each record has its own page listing every paired listing with its two monthly prices, so a reader can recompute each published per-product change and the segment median from the listed monthly prices, along with the exclusion lists and a JSON download of the segment figures. Earlier launch months appear in the all-records table on the Price Index.

The live market reading

The live reading compares each segment's median price today against its median at four earlier points: one month, three months, six months, and one year ago. The Price Index page shows the first three; the one-year window appears in the table below and on the guides. That one-year figure is the live reading, a different statistic from the frozen monthly record above, which compares each listing with itself, so the two year-over-year figures differ. Cite the record for year over year.

Every comparison uses a matched subset. A product counts toward a given window only if it has a recorded price both now and at that earlier point, and both medians are computed over that same set of products. Without this rule, a segment's figure would move whenever the catalog grew, independent of any actual price movement.

One consequence is worth stating plainly: because each window has its own matched subset, the "current" median can differ slightly between windows on the same segment. That is the fairness rule working, not an inconsistency. The index displays percentage change rather than the underlying medians for exactly this reason.

Last computed October 28, 2026 · median-based · per-period fairness subsets. The 1 year column is the live one-year reading, a different statistic from the frozen monthly record, which compares each listing with itself a year earlier, so the two year-over-year figures differ; cite the record for year over year. The record and its citation block are on the Memory Price Index.

What this data cannot tell you

This is the part most worth reading before citing anything here.

Longer windows rest on smaller, older product sets. A product counts toward a one-year comparison only if it was already being tracked a year ago. As the catalog grows, the one-year figure for a segment is computed over fewer products than the one-month figure, and it can move when that subset changes rather than when prices do. For that reason, figures over long windows should be quoted as magnitudes ("more than doubled") rather than to a decimal place. We hold our own writing to that rule.

Retail lags contract pricing. Memory is sold to large buyers on quarterly contracts, and retail prices follow with a lag of weeks. Everything on this site describes what a consumer pays at retail, which is a different series from the contract and spot prices quoted in industry reporting. Retail will show a market turn later than the industry feels it, in both directions.

A price on an Amazon.com listing is not always Amazon's price. When measured on September 8, 2026, more than two thirds of the products we track were priced from third-party marketplace offers rather than Amazon's own. The page shows the offer whose price it displays, and the stock badge describes that offer. A marketplace price can be higher, lower, or briefly absurd, and the median-based segment figures are chosen partly to limit the effect of the last case.

Flash sales and short-lived prices may be missed. At six observations a day, a price that exists for an hour may not be recorded. The history is a sample of a continuous series, not a complete record of every price the product ever carried.

Out of stock is inferred, not announced. Amazon availability comes from whether any offer exists in the data. For Newegg, the feed contains only purchasable items, so a product disappearing from the feed is how we learn it is unavailable. Both are inferences that can lag reality by hours.

The catalog is a sample. Segment figures describe the products MemRadar tracks, not the entire memory market. They are a reasonable proxy for consumer retail pricing and should be described that way rather than as the market.

How we check our own numbers

Every quantitative claim written in prose on this site is checked against the data twice: once across all products in the segment, comparing the segment's median price then and now, and once across the subset tracked through the whole window, comparing each product with itself and taking the median of those changes. A claim ships only if it holds both ways. Claims are recorded in a registry and rechecked every time the statistics recompute, so a sentence that stops being true raises a flag rather than sitting on the site.

Where a figure varies meaningfully depending on which products are counted, the prose states a magnitude rather than a precise number. Where a page displays a computed figure and also describes it in a sentence, the two are derived from the same value so they cannot disagree.

Corrections and contact

If a price looks wrong, a product is mismatched, or a figure here does not reconcile, email hello@memradar.com and it will be investigated. Corrections are made to the underlying data rather than papered over in the copy.

MemRadar is built and run by Malcolm Konner. Affiliate commissions from retailer links fund the site and never influence prices, rankings, or verdicts. The full disclosure is on the affiliate disclosure page.

Citing this data

Data on this site may be cited with attribution. For the Price Index's year-over-year figure, cite its latest frozen monthly record:

MemRadar. "Memory Price Index, September 2026 record." Frozen October 12, 2026. https://memradar.com/price-index/2026-09/

For anything else, including the live reading:

Source: MemRadar Memory Price Index, memradar.com/price-index/

If you are writing something, the data page for journalists and researchers carries the current findings as citable sentences and explains how to request a custom cut.

Figures change daily, except the Price Index's monthly records, which are frozen once computed. Every page carries the date its figures were computed; please cite the date alongside the number. If you need a cut of the data that is not published here, a specific segment, a longer window, or a product-level series, write to hello@memradar.com.

Page last regenerated: October 28, 2026. Figures on this page update automatically from tracked prices.