
Last updated: October 5, 2026 · U.S. retail prices unless otherwise stated
TL;DR — RAM prices have not turned. Between late August and late September the cheapest DDR4 kits rose about 10% while DDR5 moved in both directions and averaged out flat. The one DRAM chip price that fell drew no clear buying interest, while the mainstream DDR4 reference chip rose to a new high for this cycle. Supplier contract prices are forecast to rise for a fourth straight quarter, just more slowly each time.
RAM prices in October 2026 have not started coming down. What changed since August is which side of the market is rising.
Four of the five DDR4 kits on the most widely cited retail price index rose between August 27 and September 28, by between 8.5% and 22.8%. The ten DDR5 kits on the same index went in both directions and came out close to flat. In this one-month sample DDR4 moved considerably more than DDR5, which reverses the year-over-year pattern, where DDR5 rose far faster.
Three markets feed this report: retail kit prices, which here are desktop kits; chip prices, for chips that go into desktop, laptop and server memory alike; and contract prices, negotiated across server, PC and phone memory together. None of the three says what a used enterprise module is worth. That depends on the specific part.
Overview: are RAM prices coming down yet?
Not broadly. Retail DDR4 rose, chip prices were mixed, and contract forecasts are still positive.
| Measure | Aug → Oct | Change |
|---|---|---|
| Retail kits (floor, Aug 27 → Sep 28) | ||
| DDR4-3200 32GB | $179 → $198 | +10.6% |
| DDR4-3200 64GB | $399 → $490 | +22.8% |
| DDR5-6000 32GB | $404 → $429 | +6.2% |
| DDR5-6000 64GB | $869 → $913 | +5.1% |
| DRAM chips (spot average per chip, Aug 31 → Oct 5) | ||
| DDR4 16Gb, branded | $91.78 → $83.59 | −8.9% |
| DDR4 8Gb, branded | $44.37 → $46.32 | +4.4% |
| DDR5 16Gb, branded | $53.92 → $58.07 | +7.7% |
| Supplier contracts (quarterly forecast) | ||
| Conventional DRAM | +13–18% → +10–15% | slower, still rising |
Percentage changes are calculated from the listed values. The three groups come from different markets on different observation dates and should not be compared against each other.
Retail: DDR4 rose, DDR5 went both ways
The largest increases landed on the higher capacities. These are the cheapest listings found for each kit, not market averages, so one listing selling out can move a row — the index itself notes that exceptional deals disappear within seconds. Four of the five rose, so the 10% does not rest on the single row that jumped 22.8%: the mean is +10.2% and the median +10.6%.
| Memory kit | Aug 27 → Sep 28 | Change |
|---|---|---|
| DDR4-3200 64GB | $399 → $490 | +22.8% |
| DDR4-3600 16GB | $99 → $114 | +15.2% |
| DDR4-3200 32GB | $179 → $198 | +10.6% |
| DDR4-3200 16GB | $94 → $102 | +8.5% |
| DDR4-2666 16GB | $95 → $89 | −6.3% |
DDR5 is the messier half.
| Memory kit | Aug 27 → Sep 28 | Change |
|---|---|---|
| DDR5-5200 16GB | $239 → $293 | +22.6% |
| DDR5-6000 32GB | $404 → $429 | +6.2% |
| DDR5-5600 32GB | $389 → $409 | +5.1% |
| DDR5-6000 64GB | $869 → $913 | +5.1% |
| DDR5-6000 96GB | $1,849 → $1,899 | +2.7% |
| DDR5-5600 16GB | $214 → $214 | unchanged |
| DDR5-6000 16GB | $239 → $239 | unchanged |
| DDR5-6400 128GB | $2,339 → $2,199 | −6.0% |
| DDR5-6000 48GB | $701 → $640 | −8.7% |
| DDR5-6600 32GB | $637 → $489 | −23.2% |
The ten kits average a rise of under half a percent. That looks calm and is not: five rose, two held, three fell, and the spread runs from +22.6% to −23.2%. Part of that spread is the index showing its limits. It lists 16GB DDR5-5200 at $293 and 16GB DDR5-6000 at $239 — slower memory priced above faster memory, which happens when only a few listings are left to pick from, not because the slower kit is worth more.
A broader average across brands and tiers would show how much of the spread is real, but the most recent twelve-month comparison dates to August 17 and has not been updated. On that dataset the 64GB DDR5-6000 category averaged $1,272 against $222 a year earlier, an increase of 473% — an August figure, not a current one.
Chip prices: one fell, the rest rose
Finished kits are built from individual DRAM chips, and those chips trade on their own market — the spot market, where they change hands in the open rather than under the contracts that supply large manufacturers. Chip density is quoted in gigabits: a 16Gb chip holds 2GB of raw capacity, an 8Gb chip holds 1GB.
| DRAM chip | Spot average per chip, Aug 31 → Oct 5 | Change |
|---|---|---|
| DDR4 16Gb, branded | $91.78 → $83.59 | −8.9% |
| DDR4 8Gb, branded | $44.37 → $46.32 | +4.4% |
| DDR4 16Gb, eTT | $11.53 → $14.13 | +22.5% |
| DDR4 8Gb, eTT | $4.51 → $5.95 | +31.9% |
| DDR5 16Gb, branded | $53.92 → $58.07 | +7.7% |
| DDR5 16Gb, eTT | $23.80 → $26.30 | +10.5% |
Averages from the DRAM spot board. Percentage changes are calculated.
These are chip quotes, not module costs, and they cannot be multiplied up into the price of a finished stick. On the same service’s contract board, the chip rows come to more than the module rows for the same period. Read the spot board as a direction indicator, not a cost model.
Only one of the four DDR4 prices fell. The denser branded chip dropped 8.9%, while the 8Gb chip — the one TrendForce tracks as its mainstream DDR4 reference — rose to a new high for this cycle. Working from the weekly series, it stood at $36.00 at the end of June, so it has gained about 29% in three months.
The eTT grades rose fastest of anything on the board. eTT stands for effectively tested: chips that have been tested but are sold unmarked, or only partly marked, outside the branded channel. They are what the cheapest modules get built from. A 23% rise on the 16Gb eTT and 32% on the 8Gb means the floor under budget memory is going up.
DDR5 chips rose on both grades, 7.7% branded and 10.5% eTT. So that one DDR4 price is the only thing on the board that moved against the trend.
The weekly reports describe that price retreating without attracting clear buying interest, and on October 5 its daily quotes ran from $45.40 to $117.00 around an $83.59 average — a spread wide enough that the average describes very little. The week was also quiet through the Chinese holiday, which thins the data further.
So the DDR4 price that fell is the one with no clear buying behind it, while the chip the market treats as its reference rose. For anyone pricing modules rather than chips, that is the distinction that matters. This report flagged the easing in September; the October board shows how thin it was.
The fall does narrow August’s most striking figure. Branded DDR4 was quoted about 70% above the same-density DDR5 chip on August 31 and about 44% above it on October 5 — still inverted, but by much less.
An older generation costs more per chip than its replacement for one reason: supply is contracting faster than demand for it. DDR4 output is shrinking faster than new capacity can offset, while the installed base of DDR4 servers and workstations is still running and still needs parts.
Contract prices: a fourth straight rise forecast for conventional DRAM
Contract prices are what suppliers negotiate with their large customers, and they move slower than anything else here.
| Quarter | Conventional DRAM contract change | Status |
|---|---|---|
| Q1 2026 | +93–98% | actual |
| Q2 2026 | +58–63% | forecast; no actual published |
| Q3 2026 | +13–18% | forecast; no actual published |
| Q4 2026 | +10–15% | forecast |
The forecast has decelerated every quarter this year without once turning negative, and only the first quarter’s 93–98% is a published result — the three since are projections. DRAM industry revenue did rise 59.5% in Q2 to $154.73 billion, which is consistent with a sharp contract increase, though no contract percentage was published.
Conventional DRAM here means the commodity memory that goes into servers, PCs and phones, as distinct from the high-bandwidth memory stacked onto AI accelerators. These are quarterly changes, so they are not directly comparable with the month-over-month retail figures above.
NAND flash is projected to rise 15–20% in Q4 against conventional DRAM’s 10–15%, which matters for anyone budgeting storage alongside memory. Comparing the published ranges quarter by quarter, that puts NAND ahead again after Q2, when it was forecast at 70–75% against conventional DRAM’s 58–63%.
DDR4 contract pricing has not softened at all. The contract board shows DDR4 8Gb and 16Gb each up 4.17% in the second half of August, its most recent update. A module maker setting prices against replacement cost is still looking at an input that is going up.
Micron’s results, published September 30, point the same way. Its DRAM prices rose by a high-teens percentage from the previous quarter. Management said it has no line of sight to when supply and demand return to balance, and expects fiscal 2027 and 2028 to be much tighter than 2026. Its mobile and client unit was the one division whose bit shipments fell from the quarter before, with revenue up 14% on price and “partially offset by lower bit shipments.” That is consistent with demand resistance, buyers paying more and taking less, though product mix and supply allocation can move bit shipments too.
Where the capacity goes has not changed either. Micron’s exit from the Crucial consumer business, announced in December 2025 with consumer-channel shipments scheduled to end in February 2026, stated the priority plainly. Server memory remains the constraint: total RDIMM bit supply — how much server memory gets produced, measured in bits — is projected to grow only 15–20% in 2027, significantly lagging expected server CPU shipment growth, which means more sockets chasing proportionally less memory.
Forecasts for when this ends disagree sharply; twenty of them side by side is a better guide than any one projection.
What this month changes for decommissioned memory
Two things in this month’s data affect what to do with memory coming out of service.
The first is that old DDR4 has not got cheap. Its retail floors rose, its contract prices rose, and its mainstream chip set a new high for the cycle — while DDR4 servers and workstations are still in service and still need replacement parts. That is the opposite of the price decline that normally follows a generation change, and it is why DDR4 modules pulled from working systems are worth identifying properly instead of scrapping.
The second is the gap between chip grades, which closed. eTT 16Gb rose 23% while the branded chip fell 8.9%, so eTT moved from about an eighth of the branded price to about a sixth. That is a fact about the chip market, not about used modules — a used DIMM’s value turns on its own part number, capacity, rank, module type, condition and the demand for that specific part. A chip price board carries none of that. It does show cheap memory getting less cheap, the same direction as everything else this month.
Sorting a decommissioned lot is worth the time. What a module is worth depends on its density, its rank configuration — how many full sets of chips it presents to the memory controller — whether it carries ECC, and whether it can be identified at all. The fields on a server RAM label settle most of that, and for ECC RDIMM and LRDIMM inventory the server memory route is the specific one.
One honest limit. New-kit prices and used-module prices can move apart, and this year they have. Used 16GB DDR4 modules in Shenzhen’s Huaqiangbei market were reported down more than 30% from their January peak by late July, in a single Chinese trade report, while new retail kept climbing. A rising retail price says the replacement is expensive; it does not set what a used lot will fetch.
What would show a real turn
None of the three signs of a real turn has appeared, though one has moved.
The spot premium over contract would have to compress and stay compressed. Branded DDR4 16Gb averaged $83.59 on October 5 against a $50.00 contract price, about two-thirds above, where September was roughly three-quarters. Read that loosely. The contract board was last updated for the second half of August, so the two readings sit five weeks apart, and a premium can also compress because contract prices are rising rather than spot falling. That is not the market easing.
Contract increases would have to reach zero. The Q4 forecast is 10–15%.
And the decline would have to spread beyond the one price that fell, while the chip the market watches rose over the same weeks.
Two dates are worth watching. Samsung’s preliminary third-quarter results are due around October 8, putting a second supplier’s pricing beside Micron’s. And the spot reports after Chinese holiday trading resumes will show whether the DDR4 16Gb retreat was a correction or a thin week with nobody bidding.
Until prices fall on the chips that actually trade, slower increases are still increases.
Methodology
Four pricing layers, kept separate:
- Retail floor — the lowest available price for a complete new RAM kit tracked among major U.S. retailers, in GB of total kit capacity. Desktop kits.
- Average retail price — an approximate category average across kits sharing a capacity and speed class. The most recent dataset is August 17, 2026.
- DRAM spot price — open-market pricing for individual DRAM chips, quoted in Gb of chip density. Not a finished-module price, and not multipliable into one.
- DRAM contract price — negotiated pricing between suppliers and large buyers, reported as quarterly change.
All figures are third-party and published: retail prices from a public price tracker, chip spot and contract prices from market-research boards, forecasts from market-research releases, and supplier figures from company disclosures. Each is linked at the figure it supports so it can be checked against the original.
The report also crosses two memory segments. PC and client memory covers desktops and laptops, typically on UDIMMs or SODIMMs, and the retail tables here are desktop kits. Server memory is a separate enterprise market, commonly ECC RDIMMs or LRDIMMs, with its own qualification, procurement and pricing. Chip spot prices and conventional-DRAM contract prices sit upstream of both.
Observation dates differ by measure: retail floors are August 27 and September 28, spot averages are August 31 and October 5, contract figures are quarterly forecasts. Month-over-month, premium and ratio figures are calculated from the listed values.
The retail floor index records the single cheapest listing per kit, so rows are sensitive to one listing appearing or selling out. Where it contradicts itself — a slower kit priced above a faster one — that is noted rather than smoothed.
This report is not a used-memory valuation index. Secondary-market value depends on the specific module, capacity, part number, quantity, condition and current demand.