We do not predict prices. On 2026-09-22 new DDR5 desktop kits start at $8.63 per GB and DDR4 at $4.43; the two suppliers' statements differ and are shown side by side. Buy the capacity you need now, do not buy speculatively, and a working kit you are happy with is not a reason to buy.
Live prices updated · refreshed through the day · outlook reviewed monthly
Buy the capacity you need now, do not buy speculatively, and a working kit you are happy with is not a reason to buy. Part of this market's demand is fear of higher prices, and buying on fear is how a buyer ends up owning capacity they did not need at the top of a price cycle. If your machine works and the memory in it is enough for what you run, the right amount to buy today is none.
Every figure in this section is read from our own catalogue when the page loads, and it is a LEVELS observation, not a rate. The floor is the single cheapest new kit per gigabyte; it can move because one cheap kit appeared or sold out with no seller changing a price, which is why percentage changes on this site come from a matched-model index that compares each kit only with its own earlier price. As of 2026-09-22 the cheapest new DDR5 desktop kit is $8.63 per gigabyte across 210 in stock (median $17.81), and the cheapest new DDR4 desktop kit is $4.43 per gigabyte across 71 in stock (median $8.01).
Floors grow and shrink with catalogue membership, and this catalogue was built recently: the seed run that created it is the reason a kit can appear here that was not tracked last week. That is a property of how a young series is assembled, not of the market, and it is the reason no direction is read from a floor on this page.
Tracking since 2026-09-22. No complete day yet. A line needs two complete days, and nothing before our record is reconstructed, so no chart is drawn yet.
Tracking since 2026-09-22. No complete day yet. A line needs two complete days, and nothing before our record is reconstructed, so no chart is drawn yet.
Our record for these kits starts on 2026-09-22. Nothing before it is backfilled or reconstructed, and the long-run story on this page comes from named, dated reporting from other people, labelled as theirs and set beside ours. The charts below draw complete days only and end at the last complete day, because the daily snapshot is taken once a day while the live figures above are read at page load, and mixing the two once put a partial day beside a live figure on the same screen.
A matched-model index needs two complete days before it can print a reading, and it prints the number of kits it compared. Until it does, this page says so instead of offering a direction. An outlook that pretended to a trend from a single observation would be the kind of claim we set out to avoid.
| Whose figure | What it is | Value / date |
|---|---|---|
| MEASURED BY US | Cheapest new DDR5 desktop kit, per GB - 210 kits, a floor not a market rate | $8.63/GB, 2026-09-22 |
| MEASURED BY US | Cheapest new DDR4 desktop kit, per GB - 71 kits | $4.43/GB, 2026-09-22 |
| REPORTED BY ZAK KILLIAN | PCPartPicker average prices, as pulled by Tom's Hardware: a 5600 MT/s 2x32GB (64GB) DDR5 kit averaged $191 in August 2025 and $1,118 in August 2026, +485%; a 6000 MT/s 2x32GB kit went from $222 to $1,272, +473%. Tom's calls the data 'somewhat approximate' but 'broadly accurate'. | |
| REPORTED BY 3DCENTER INDEX | 3DCenter's index of German retail DDR5 prices (Geizhals data, July 2025 = 100%) stood at 544% in September 2026, up from 486% one month earlier. | |
| REPORTED BY TRENDFORCE DRAMEXCHANGE SPOT PRICING | The DDR4 1Gx8 3200 (8Gb) DRAM chip reached a record $42.50 in spot pricing on 10 August 2026, the highest since DRAMeXchange began tracking it in June 2016. |
Memory is the category where the supply story is most concentrated and best documented. Three manufacturers control over 95% of global DRAM output, which means the market has no long tail to absorb a shock - what those three decide to make is what exists.
What they decided to make is High Bandwidth Memory for AI accelerators. HBM took roughly 23% of DRAM wafers in 2026 against about 8% in 2024, and a wafer producing HBM is a wafer not producing the conventional DDR4 and DDR5 that servers use.
The economics behind that choice are not marginal. HBM has been reported to earn three to five times the revenue per wafer of conventional DRAM, which makes the allocation decision straightforward for a manufacturer and unfavourable for everyone buying ordinary server memory.
And the decision is already made for this cycle: SK hynix has confirmed its entire 2026 HBM production is committed. Capacity that is committed is not capacity that responds to conventional demand, whatever conventional demand does.
The paragraphs above are the same supply-side facts, from the locked citation list, that the server RAM forecast reads - they are imported, not copied, so the two pages cannot drift. Consumer and server memory are downstream of one wafer-allocation decision, which is why they share this section and differ everywhere else.
Everything in this box is REPORTED by the named source and read by us at that source on 21 September 2026. None of it is our measurement, and it is never averaged with ours. Where a source's evidence chain is weaker than we would like, the note says so.
Bad bots account for 91% of the traffic reaching one retailer's DDR5 product pages, about ten automated requests per legitimate visit (DataDome's Jerome Segura, in a LinkedIn post). Newegg and Amazon have run repeated CPU-plus-memory bundles that price a 32GB kit well below its standalone cost; Micro Center sells its CPU, motherboard and 32GB DDR5 bundles in-store only, one per household; Framework stopped selling standalone RAM in November 2025, citing scalpers.
Source: Luke James, Tom's Hardware, citing DataDome. read the source
Two memory makers have said publicly, in the past three months, how long they expect the constraint to last, and their statements differ in kind. SK hynix's chief executive, speaking to Reuters on 10 July 2026, said demand would stay above the company's supply capacity even beyond 2030 and that 2027 would be the worst year in the industry's history from the supply perspective. Micron's chief executive, on the company's earnings call of 24 June 2026, said industry supply would improve gradually in 2028, and also that Micron has no line of sight as to when memory supply will catch up with demand.
Those two are not the same forecast and they should not be averaged into a single date. One is a statement about demand exceeding one company's capacity for years; the other is about when supply starts to improve, with an explicit admission that catching up has no date. A midpoint of 'sometime around 2029' would be a number neither company said. Both come from suppliers with a commercial interest in how the market reads them, which Tom's Hardware notes about the first, and both are reported through the outlets shown below with their caveats.
What can be said honestly is what they have in common: neither expects relief in the near term, and the earliest improvement either names is 2028. Beyond that we do not extrapolate, which is also why this page has no target price and no date for a fall.
Everything in this box is REPORTED by the named source and read by us at that source on 21 September 2026. None of it is our measurement, and it is never averaged with ours. Where a source's evidence chain is weaker than we would like, the note says so.
SK hynix CEO Kwak Noh-jung told Reuters: 'We forecast that next year will be the worst year in the industry's history from the supply perspective' and 'We still forecast that customer demand will remain higher than our supply capacity even beyond 2030.'
Source: Kwak Noh-jung, SK hynix CEO, to Reuters on 10 July 2026, as reported by Tom's Hardware (Jake Roach). read the source Note: The statement is about demand exceeding SK hynix's supply capacity, made on the day the company listed on Nasdaq. Tom's Hardware notes forecasts like this are 'tricky' coming from a supplier. The TechPowerUp article named in the brief returned HTTP 403 to two readers and was not used.
Micron CEO Sanjay Mehrotra on the fiscal Q3 2026 earnings call: 'Our customers are recognizing that supply shortages in memory and storage will take considerable time to improve, even as we expect industry supply to improve gradually in 2028,' adding 'we currently do not have line of sight as to when memory supply will be able to catch up with increasing demand.'
Source: Sanjay Mehrotra, Micron CEO, earnings call of 24 June 2026, as reported by CNBC and TechSpot. read the source Note: Two independent outlets carry the same quotation (CNBC, 24 June 2026: https://www.cnbc.com/2026/06/24/micron-mu-earnings-report-q3-2026.html). Micron's own transcript page did not render, so the wording is as reported, not read from Micron's IR site.
Directionally: on the supply facts we have cited, the pressure on consumer memory prices remains upward, and the earliest improvement either supplier has named is 2028. That is a statement about direction and about the sources, not a price or a date. We do not print either.
Consumer retail is the part of the market where timing is least predictable. Server contracts reprice on a quarterly cadence and move first; consumer retail is cushioned by inventory bought at older prices and then moves suddenly when that runs out. Prices that stop rising after several-fold increases have not fallen, only stopped rising, and weekly coverage routinely mistakes the two, so a flat month should not be read as the start of a decline.
Where the honest answer is unclear we print unclear. It is unclear how far buyers deferring purchases can offset constrained supply, and it is unclear whether DDR4 and DDR5 will behave alike from here, which is what the matched-capacity comparison exists to keep testing.
An outlook with no stated falsifiers cannot be wrong, and a claim that cannot be wrong is not information. The first four are the supply-side falsifiers shared with the server memory forecast; the last three are specific to consumer retail.
We check these monthly and publish the result as a dated check-in whether or not the read changed. An outlook that never revises is not being tested.
If your machine works, do not buy. If it needs memory to run, buy the capacity it needs, compare per gigabyte at that size, and buy from the maker's storefront or an authorised seller, because in a shortage the price that looks too good is usually information about the seller. Do not buy capacity you will not use because of a headline about prices: the same reporting that tells you prices may rise tells you nothing that is a reason to own idle memory. Buy the capacity you need now, do not buy speculatively, and a working kit you are happy with is not a reason to buy.
Every monthly edition, newest first, archived rather than overwritten. Each one opens by stating what the previous edition said and what our data actually did - an archive that never grades itself is a blog, not a record.
Route note: this is a sibling of the server RAM forecast rather than a section of it. 8 pages are registered in the /ram cluster; 5 verified additions from the locked list also inform the server RAM forecast.
We do not predict prices. On the supply facts we have cited the pressure remains upward, and the earliest improvement either supplier has named is 2028. SK hynix's CEO said demand would stay above its supply capacity even beyond 2030; Micron's said supply would improve gradually in 2028 with no line of sight to demand being met. Those are printed side by side, not averaged. Buy the capacity you need now, do not buy speculatively, and a working kit you are happy with is not a reason to buy.
No supplier has given a date for prices to fall, and we will not invent one. Micron said industry supply improves gradually in 2028; SK hynix said demand stays above its supply capacity even beyond 2030. Those are different statements from interested parties. The falsifiers section says what would change the outlook.
Our record for consumer kits starts on 2026-09-22, so we do not yet print a direction from it; the matched-model index needs two complete days before it can. What we can say is where the cheap end sits today, and that a floor is a level rather than a rate.
DDR4 shares the same wafers and the same three makers as DDR5, and reported DDR4 chip spot prices reached a record in August 2026. Whether DDR4 and DDR5 diverge from here is what our matched-capacity comparison tests; we do not forecast it.
If your machine works, no: Buy the capacity you need now, do not buy speculatively, and a working kit you are happy with is not a reason to buy. If it needs memory to run, buy the capacity it needs and compare per gigabyte at that size.
Because the three companies that make nearly all DRAM have pointed their best capacity at high-bandwidth memory for AI accelerators, which takes about 23% of DRAM wafers in 2026 against roughly 8% in 2024 and earns several times the revenue per wafer. The mechanism is set out in our server RAM shortage page and the why-RAM-is-expensive article.
A sustained fall in our matched-model index, HBM's share of wafers falling back, committed 2026 capacity being reallocated, bot-driven buying ending, or a dated consumer restock. The falsifiers section lists each, and we check them monthly.
Live figures refresh through the day; the outlook is reviewed monthly and published as a dated check-in, and each edition opens by grading what the last one said against what our data did.