RAM Prices: What Desktop and Laptop Memory Costs per GB Today

How to read a RAM price

RAM is sold in three shapes at once, and mixing them up is the commonest way to misread a price. A module is one stick. A kit is a matched set of two or four sticks sold together, so that a dual-channel board can use them as a pair. And a listing's headline capacity is the total of the kit, which means a '32GB' kit is normally two 16GB modules. Every price on this page is divided by the total kit capacity, so a two-stick 32GB kit and a single 32GB stick land on the same footing per gigabyte, and the table shows the module count beside the total so you can see which you are looking at.

Cost per gigabyte is the only figure that lets you compare a 32GB kit with a 64GB kit on the same terms, and in this market it is also the figure that most often surprises people. The larger kit is not automatically the cheaper gigabyte. Sometimes it is, sometimes the ladder bends the other way, and the capacity pages linked below compute it live rather than assuming it. What the per-gigabyte figure cannot tell you is whether you need the capacity at all, which is a separate question and the one the rest of this page keeps returning to.

The rated speed and latency columns are the maker's stated specification, printed exactly as the listing gives them. We do not rank by them, we do not say a kit is quicker or slower than another, and we do not claim any speed is worth paying extra for. The one thing that is a matter of fact rather than preference is compatibility: DDR4 and DDR5 are physically keyed differently and a board takes one or the other, so the generation column decides what you can use before any other column matters.

What is driving the price, and where we stand on it

The mechanism behind this market is documented elsewhere on this site and we do not restate it here. The short form is that the three companies that make nearly all of the world's DRAM have pointed their best capacity at high-bandwidth memory for AI accelerators, leaving less for the commodity modules that go in a desktop. The plain-language version, with the wafer figures, is in why RAM is so expensive in 2026, and the server-memory view of the same shortage, measured on our own catalogue, is on the server RAM shortage page.

What is new on this page is the consumer side measured directly. Until now the honest answer to 'what does a desktop kit cost' on this site was 'look at the reported figures', because our RAM tracking was ECC server memory and said so. This section is the other half. As of 2026-09-22 the cheapest new DDR5 desktop kit we track costs $8.63 per gigabyte, and the cheapest new DDR4 desktop kit $4.43 per gigabyte. Those are floors, single cheapest listings, and they are labelled that way everywhere they appear.

On outlook, two suppliers have said publicly how long they expect the constraint to last, and they said different things. SK hynix's chief executive said demand would stay above its supply capacity even beyond 2030. Micron's said industry supply would improve gradually in 2028 while adding that it has no line of sight to demand being met. Those are not the same kind of statement, they come from companies with an interest in how the market reads them, and we print them side by side rather than averaging them. The full treatment, with what would change the outlook, is in our RAM price forecast.

Rush buying, bots and bundles: why the shelf price is not the whole story

Part of what buyers see on a listing is not the price of memory but the behaviour of the market around it. Retail trade reporting from August 2026, read at source and set out in the box further down, describes automated traffic making up the large majority of visits to one retailer's DDR5 product pages, kits being tied to processors in bundles that price the memory below its standalone cost, and at least one PC maker no longer selling standalone RAM at all. None of that is our measurement and we attribute all of it. What it means for a buyer is narrow and practical.

First, a low price on a single listing is not always an offer you can act on. Cheap kits in a shortage tend to vanish within minutes, which is why our tables show only what is in stock at the moment of the last check and why the timestamp on this page matters. Second, a bundle can make memory look cheap by attaching it to a purchase you did not plan to make. If you need the processor or the board anyway, that can be a real saving. If you do not, the bundle is a way of spending more to feel you saved. Third, an unusually low standalone price in this market is a reason to slow down, not speed up, for reasons the counterfeit section below sets out.

Which memory is which: UDIMM, SO-DIMM and RDIMM

Consumer desktops use unbuffered DIMMs, universally shortened to UDIMM. Laptops and small-form-factor machines use the shorter SO-DIMM. Servers and many workstations use registered modules, RDIMM, which add a register chip between the memory controller and the DRAM and are not electrically the same product as a desktop stick. The three are sold under the same generation names, DDR4 and DDR5, which is exactly why they get confused.

The practical rule is that a desktop board built for unbuffered memory will generally refuse a registered module, and a server board built for registered memory will generally refuse a plain desktop stick. The reasons, and the exceptions, are set out in RDIMM vs UDIMM, and we do not repeat them here. Whether a particular machine takes what you have in mind is answered by that machine's own manual or the maker's specification page, not by a generation label on a listing and not by us.

ECC deserves one sentence of its own because DDR5 muddied it. Every DDR5 module carries on-die error correction as part of how the chip works, which is not the same thing as system-level ECC, where the platform checks and corrects data across the whole memory path. Unbuffered ECC modules exist and we record them, but they are a separate class that consumer boards mostly do not support, so they are kept out of every figure on these pages.

Why some kits are so cheap, and why that is the thing to be careful about

In an ordinary market a price far below everyone else's is a bargain. In a shortage it is a signal, and usually not a good one. When genuine kits are scarce and expensive, the incentive to sell something that only looks like a kit rises with the price, and reporting since spring 2026 has described several different versions. Some are openly what they say they are and simply not what a buyer expects: V-Color sells DDR5 filler modules that contain no memory at all and are labelled as decorative, and they appear on retail sites next to real RAM. Some are alleged fraud: posters in Japan described laptop DDR5 modules carrying plastic dummy chips glued to the board, and a buyer described a module sold as Kingston FURY DDR5 that, on teardown, held DDR4 chips. The sourced box below carries each with its date and its caveats.

We handle this at the door rather than at the table. A listing enters our catalogue only if the brand opens its title and Amazon's own brand field agrees, the capacity in the title matches the module count and the per-module size, the feature bullets do not contradict the title, and nothing in the listing says dummy, filler, decorative or zero gigabytes. Then, for every generation, form factor and capacity, we compute a plausibility floor from the established-brand listings that passed those gates: forty-five percent of their median price per gigabyte. Anything materially below it is rejected and logged instead of shown. The floor is computed, never hardcoded, so it moves as the market does.

The live floors are in the table below. A kit priced under its row is not proven fake, and a kit above it is not proven genuine; the floor is a screening tool for our catalogue and a reason for you to look harder at any listing, anywhere, that prices well beneath the market. The practical defences are the same as ever: buy from the brand's own storefront or an authorised seller, check that the module label and the listing agree, and treat a price that looks too good as information about the seller. The reject log from our seed run, by pattern, is summarised in the method section.

How this catalogue is built, and what it does not include

We track Amazon US listings only, in stock at the time of the last check. Prices refresh through the day, the newest observation sets the timestamp at the top of this page, and a kit that has not been re-checked in seven days is not shown. The brand list is strict: Corsair, G.Skill, Kingston, Crucial, Micron, TeamGroup, Patriot, Samsung, SK hynix, ADATA and XPG, PNY, V-Color and Lexar. Any other name has to come with a maker specification page we can read, and none has yet, so listings under other names are rejected rather than shown with a caveat. In the seed run of 2026-09-22, 530 listings reached the title gates, 520 were verified in stock against a live offer, 520 were kept and 319 were rejected across every stage. The largest single reason was a brand outside the strict list or not at the front of the title (209 of the 319).

Two things are deliberately not here. There is no performance ranking: no kit is recommended for a game, a creative workload or a speed tier, and none is called worth a premium on speed, because a listing's stated speed is a specification and what it does in your machine depends on your machine. And there is no price history before we started: our record begins on the date shown in the index above and nothing earlier is reconstructed or backfilled. The long-run story on these pages comes from named, dated reporting from other people, labelled as theirs, next to our own record, labelled as ours.

Frequently asked questions