Server RAM Price Forecast - What the Data Says

Cheapest in stock right now

As of 2026-09-29, 12% of scored server memory carry a GOOD PRICE or GREAT PRICE Historical Value signal. The full distribution is 1 great, 3 good, 10 fair, and 20 high. The remaining 12 read In progress: too few days of records, or a price that has not moved enough across its record to call low or high. These are shared Good Price status counts, not displayed historical prices.

In-stock only, cheapest per gigabyte first.

Where server memory prices are right now

Every figure in this section is read from our own catalogue at the moment you loaded the page, and it is a LEVELS observation rather than a rate. The floor is the single cheapest in-stock listing per gigabyte; it can move because one cheap listing appeared or vanished, without the market having done anything. That is why percentage changes on this site come from a matched-model index that compares each listing against its own prior price, and never from comparing one day's floor to another's.

As of 2026-09-29, the floor sits at $4.609/GB and the median across 46 in-stock listings at $12.25/GB. The gap between those two is worth as much as either number on its own: a floor far below the median means the cheap end is one or two listings deep, and a floor close to it means the whole catalogue is priced tightly together.

The two condition tracks price separately and we never merge them. New stock starts at $5.436/GB and the refurbished track at $4.609/GB. A refurbished floor well below the new one is the ordinary state of this market rather than a bargain signal, and scoring one against the other would manufacture a discount that does not exist.

The picks table further down carries the same data at row level, in stock only, so a reader who decides to act can do so without leaving the page. It never shows an out-of-stock row, which is why you will not find an empty price cell anywhere in this cluster.

The trend, on an honest window

Our snapshots for this category begin on 2026-03-26 and run to 2026-09-28, which is 187 days of daily observations. That window is stated because a trend without its window is not checkable, and because 187 days is a real but modest history - long enough to show a level, short enough that a single unusual month shapes it.

Over the most recent 30 days of that series - 2026-08-29 to 2026-09-28 - the daily floor moved from $4.468/GB to $4.468/GB, a change of 0.0%. Read that as what the cheapest listing did, not as what the market did.

The chart is a levels series by design. A levels chart answers 'what does the cheap end cost' and it answers it honestly; it does not answer 'has the market moved', because a single listing entering or leaving changes it. Both questions are legitimate and they need different instruments, so we do not use one chart to imply the other.

The supply side

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 outlook, as of 2026-09-29

Directionally, as of the date above: this is the clearest of the three categories, and the read is that pressure remains upward. Wafer allocation has moved, the economics driving it are large rather than marginal, and the 2026 supply is already committed.

The distinguishing feature of this category is that the constraint has a stated horizon. A committed production year is a fact with an end date attached, which is more than either of the other two categories offers. It does not tell us what happens after, and we will not extrapolate past what has been confirmed.

For buyers this makes the memory market the one where the supply story most directly explains the price you are seeing, and the one where waiting has the weakest case on the available facts.

Why is RAM so expensive right now?

Because the three companies that make almost all of the world's DRAM have pointed their best capacity at something else. HBM - the stacked memory that sits beside an AI accelerator - consumes roughly 23% of DRAM wafer supply in 2026 against about 8% in 2024, and it earns three to five times the revenue per wafer that conventional DRAM does. A manufacturer facing that choice does not need to be gouging anyone to produce this outcome; it needs only to allocate rationally. The DRAM market is more than 95% concentrated across three makers, so there is no fourth supplier for the displaced demand to go to.

The consumer side of that squeeze has been severe and it has been reported in figures we can point at rather than describe. DDR5-5600 2x32GB kits that averaged a little over $200 for much of the preceding eighteen months rose to around $800 between October and December 2025 on PCPartPicker's average-price data, with DDR5-6000 kits at roughly $900 over the same window. MSI told investors 2026 would be the most challenging year since the company was founded, on memory and GPU costs, and signalled hardware price rises of 15-30%.

On the contract side, TrendForce's own 1Q26 outlook was revised UPWARD rather than down - conventional DRAM from a previous estimate of +55-60% quarter-on-quarter to +90-95%, published on 2 February 2026. We cite the revision rather than the widely-quoted 55-60% because the body that produced the first number withdrew it.

We set the whole mechanism out in plain language, with the consumer anchors and our own measured index, in why RAM is so expensive in 2026.

And the people closest to the supply say it is structural. SK Group chairman Chey Tae-won said in March 2026 that the shortage is likely to persist through 2030, with wafer supply trailing demand by more than 20% and a four-to-five year lead time on the capacity that would close the gap.

Are RAM prices dropping?

Not in our data, and we are in an unusual position to answer this rather than guess at it. Our own floor and index for ECC server memory are on this page with their window stated, and the chart is a levels series - it shows what the cheap end costs rather than asserting a direction. Percentage movements on this site come from a matched-model index that compares each listing against its own prior price, because comparing one month's median to another's measures which listings we added, not what sellers did.

There is a version of the falling-prices story circulating that rests on retail kits stabilising at their elevated level. Stabilising is not falling. Prices that stop climbing after quadrupling have not come down; they have stopped going up, and the two get conflated constantly in coverage because the weekly delta is what gets reported.

What would genuinely count as prices dropping is set out in the falsifiers section below, and none of it has fired. The most important single item is HBM's share of wafer supply falling back toward its earlier level, because that is the mechanism that displaced conventional DRAM in the first place.

DDR5 price forecast, and the DDR4 story underneath it

The DDR4 half of this market is the more interesting one and it is the half almost nobody measures. When manufacturers moved capacity to DDR5 and HBM, DDR4 became a legacy product still needed by an enormous installed base of servers and desktops that cannot use anything else. A shrinking supply meeting inelastic demand is the textbook setup for a price rise, and DDR4 has been rising faster in relative terms than DDR5 in much of the reporting - to the point that MSI is reported to be shifting its motherboard mix back toward DDR4 boards.

The table above is our own measurement of that split, computed live. It is ECC registered server memory rather than consumer kits - desktop and laptop kits are measured separately in the RAM section - which matters for the absolute numbers and not for the direction. As far as we can establish it is one of very few continuously tracked DDR4 series published anywhere - most DDR4 coverage cites a retail anecdote or a screenshot of a tracker graph, and we would rather show a series with its start date attached.

On DDR5 specifically we will not print a target price or a date, for the same reason we do not in any other category. What we will say directionally is that DDR5 sits closest to the HBM displacement, since it shares the advanced process nodes being reallocated, and that no reported figure we can find describes that reallocation reversing.

What we track, and what we do not

This scope note exists because the question that brings most people here is about a desktop RAM kit and the data on this page is about server memory. This page tracks ECC registered and load-reduced server memory - the modules that go into rack servers - priced per gigabyte from live in-stock listings. Desktop kits and laptop SO-DIMM are tracked too, but separately: their live prices are in the RAM section and their own outlook, with the same falsifiers, is on the consumer RAM forecast. No figure on THIS page should be read as the price of a gaming kit.

Any consumer figure that appears on this page is attributed reporting, labelled as reported in the anchors table, and carries its source; our own measured consumer figures live on the RAM pages and are never mixed into this page's rows. Every measured figure here is ours and is labelled measured. The two never share a row and we do not average across them.

What justifies putting them on the same page is that the mechanism is shared and sourced: the same three manufacturers, the same wafer allocation decision, the same HBM displacement. Server memory and consumer memory are downstream of one supply decision, which is why the server index measured here and the consumer index measured in the RAM section are two readings of the same supply decision - different points on the same curve, not different curves.

What would change this outlook

This is the section that makes the rest of the page worth citing. An outlook with no stated falsifiers cannot be wrong, and a claim that cannot be wrong is not information. Here is exactly what we are watching, and what each of these moving would mean:

HBM's share of DRAM wafers falling back toward its earlier level. The 23%-versus-8% comparison is the load-bearing figure here, and a reversal would show up in the same reporting.

The revenue-per-wafer gap narrowing. The three-to-five-times economics are what make the allocation rational; if that compresses, conventional DRAM becomes worth making again.

Committed 2026 capacity being reallocated or expanded - a manufacturer stating that conventional DRAM output is rising would directly contradict the current read.

Our own index establishing a sustained downward level. The RAM price index on this site is the check on all of the above, and it is built to show levels rather than to flatter a narrative.

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.

What a buyer does with this

If a server needs memory, buy the memory. This is the category where the supply facts most clearly explain the price, and where the case for waiting is weakest on the evidence available.

Buy on cost per gigabyte and check the used and pulled-module market, which prices partly independently of the new-module constraint - our memory pages carry both.

If you are specifying a new build, size the memory deliberately rather than generously. In a constrained market, the cheapest gigabyte is the one you correctly decided not to buy.

Where to go next

The main forecast page covers the method and the hard drive picture in full · what counts as a good price · the cheapest per terabyte right now · the live hard drive market · hard drive forecast · SSD forecast · consumer RAM forecast · PS5 SSD forecast · external drive forecast · used drive forecast · the storage crisis tracker follows the category-level picture.

Frequently asked questions