RAM Prices in 2026: Buy Now or Wait? What Waiting Has Actually Cost

Everybody answering the RAM prices question — buy now or wait — answers it by predicting, and the predictions in circulation for 2026 disagree with each other by more than the sums involved. The stakes are real: a 32 GB DDR5 kit that sold for around $110 in 2025 now runs several hundred dollars, so getting this wrong by a month costs more than a graphics card used to. This page answers the half that can be checked, from our own dated observations: what waiting has cost so far, kit by kit, against the money sitting on the shelf today. We measured every tracked part from the first observation in our window to the latest, and read the pages ranking above us to see what they charge the question with.

One of them turned out to be reporting a vanished promotional listing as a 95% market move on a drive whose real series we hold, 44 observations deep. Start with what waiting cost.

What the last 90 days of waiting actually cost

Our series run on a rolling 90-day window, which is short, and we would rather say so than dress it up. What it holds is every observation we recorded on a named kit at a named retailer, so “what would waiting have cost me?” has an arithmetic answer for each one: the first day of the window against the last.

The median memory move across those 10 kits is +9.2%. That is the number the timing question is really being asked about, and it is worth holding next to the headline percentages further down this page, which run into the hundreds. It is also an average over kits that did different things: 7 rose, 1 fell and 2 held flat.

Storage did something different over the same days, from the same scrapes.

Median storage move: +1.0% across 7 drives, with 4 up and 3 down. Both medians describe the same weeks and the same retailers. If somebody tells you what “prices” did this quarter without saying which of these two markets they mean, they are describing a mood.

The clean experiment: the cheapest part changed identity

The strongest thing a young series can say about timing needs no forecast at all. Inside a single capacity band, where every part does the same job in the same slot, the price order can invert — the drive that was the expensive one becomes the cheap one. When that happens, “wait for the price to come down” and “buy the cheapest one” stop being the same instruction, and only the second one was answerable on the day.

It is also the cleanest answer to the “rising or falling” framing: on our observations it did both at once, inside one product class.

The spread you can capture today against the drift you cannot predict

Every page arguing about timing is arguing about the number in the last section. Almost none prices the number in this one: what the cheapest and the dearest listing in a band costright now. That figure needs no waiting, no forecast and no luck.

So the comparison that matters comes out like this: the widest same-day spread we can see is 33.3% ($80.00 on 32 GB DDR4), against a median 90-day drift of +9.2%. Choosing the right part inside one band is worth 3.6x the whole quarter’s typical movement — and unlike the movement, it is a decision you get to make.

Cross-retailer shopping is the weaker version of the same lever. Where we see one part in stock at both Newegg and B&H the gap has been small — far smaller than the gap between two different parts of identical specification. The money is in which part, not which shop; the brand comparison finds the same thing with capacity, speed and CAS latency all held still.

The dip that existed, and the day it closed

“Wait for a dip” is the most common advice on this question and the hardest to act on, because a dip is only visible once it has gone. Our window shows several, and it also shows what happened next.

The deepest one bottomed at $249.99 on Sep 7 and sits at $319.99 now. A buyer who was watching that part got 28.6% off the window’s opening price; a buyer who waited for the next one is still waiting. The other end is worth the same attention: SK hynix Platinum P41 2TB NVMe Gen4 SSD peaked 66.0% above its opening price on Sep 30. Waiting is not a one-way bet on the way down — it is exposure, and some of these series spent weeks above where they started.

A threshold is the only mechanical way to act on a dip you cannot see coming, and it has floors underneath it that most alert tools never mention — the numbers a RAM alert actually needs.

A delisted deal is not a market move

Here is where the published advice goes wrong in a way you can check. Several of the pages ranking for this question draw their price history from listings that no longer exist — a promotional bundle, a clearance SKU, a marketplace seller — and report the disappearance as though the market had repriced by the same amount.

Take the clearest case. bottleneckpc.com, updated 18 September 2026, tells readers that “the Samsung 990 Pro 2TB we caught at $200 in July now sits near $390”. Its own body text is more careful: the $199.95 figure was a heatsink bundle listing, since replaced at $499.95, while “the bare drive floats around $385-390”. Both halves are probably true. But the FAQ version reads as a 95% move on a named drive, and our series for that exact SKU — the deepest we hold — shows nothing like it.

The same page lists the SK hynix Platinum P41 2TB at “~$360 where listed” on an 18 September update. Our observation for that drive is materially higher and rising through the window — both are in the storage table above. A price printed without its observation date is not a stale sentence, it is a wrong answer: the same failure we found in the RAM versus SSD comparisons, which priced memory at 2026 levels and storage at 2025 levels in one article.

What the forecasts say, and what they are worth

We do not publish forecasts. That is a rule this site was built on, not a gap in the data, and this is the query where it costs us the most traffic, so it is worth explaining. A forecast is an unfalsifiable input to a decision you make today; an observed price is a falsifiable one. We would rather be the site that hands you the second kind.

Read down that table and the range is the finding. The contract figures are large, sourced and real — and they describe hyperscaler procurement of server parts, not a kit at a US retailer, which is the commonest confusion in this year’s memory coverage. The retail forecasts are the ones aimed at you, and the ones without a named basket. Our measured retail moves, above, are an order of magnitude smaller.

Why the upstream numbers do not reach the shelf at the same size is a longer story about wafer allocation and AI demand: why RAM prices are so high.

Five rules that work without a forecast

None of these needs to know where prices go next, which is the point of them.

What these numbers are, and what they cannot tell you

Every price above is an in-stock observation from a live retailer product page, with the retailer and the date attached. Out-of-stock listings record no price rather than carrying the last one forward, so a “move” here is never an artefact of a listing going dark. A jump over 60% is held until a second scrape confirms it, and demo rows are blocked from public pages at the database layer.

The limits matter as much. Our window is 90 days on a series that began 22 June 2026, so nothing here speaks to 2025 — the large year-on-year multiples in circulation describe a period we did not observe. Our basket is around 20 high-liquidity parts at US retail, and says nothing about a part we do not track. Scrape cadence runs from 48 to 240 hours per listing, so a short-lived price can fall between two observations. And where a category sits below our five-part floor we withhold the aggregate rather than publish a thin one. None of that is fixable by writing a forecast instead.

Related reading

FAQ

Context: DDR5 from $16.53/GB, NVMe Gen4 from $0.184/GB, cheapest in-stock observations in the last 30 days.