Why Are RAM Prices So High? What Our Tracked Data Actually Shows

A 32GB DDR5 kit that cost about $80 two years ago will now clear $500 at a mainstream US retailer. If you are speccing a fleet refresh, that single line item can eat the margin you planned the whole quarter around — and the quotes you collected three weeks ago are already stale.

So: why are RAM prices so high? The honest answer has two halves. One half is a supply story you can read anywhere. The other half is what the prices are actually doing this week, which almost nobody publishes, because it needs somebody to go and look every day.

We do that looking. ByteWatch scrapes real listings at Newegg and B&H, records every observation, and computes a price-per-gigabyte index from them. Every figure below is rendered live from that data at the moment this page was built — not typed into the copy by a writer. Where our data is too thin to support a claim, we say so instead of rounding up to a story.

The short answer, in one number

RAM is expensive in a way that storage simply is not, and the cleanest way to see it is to stop comparing kits and start comparing gigabytes. A gigabyte is a gigabyte. Here is what one costs, per category, across every SKU we currently track:

That multiple is the whole story in one figure. Nobody is shocked that flash is cheaper than memory — they are different technologies doing different jobs. What is remarkable is the size of the gap, and that it has been widening while flash sat still.

What we actually observe

We index each category by taking, for every SKU, the cheapest in-stock price across the retailers we track, dividing by capacity, and taking the median across SKUs. The series starts at 100. Every per-kit series behind these numbers is plotted on our live RAM price chart. Here is where each category sits against its own starting point:

Why DDR5 went up: HBM ate the wafers

The mechanism is capacity allocation, and it is not mysterious. High-bandwidth memory — the stacked DRAM bolted to AI accelerators — consumes roughly three to four times the wafer area per gigabyte that conventional DDR5 does. It also sells for vastly more per unit. Faced with a fixed number of wafer starts and two products to make from them, Samsung, SK hynix, and Micron did the arithmetic every manufacturer would do and pointed their lines at the expensive one.

Consumer and channel DDR5 did not get cancelled. It got the residue. That is why the price moved without any change in what you are buying: the kit on the shelf is the same kit, made on a line that now has better things to do.

The supply side confirms it. TrendForce reported on 9 July 2026 that RDIMM bit supply will grow only 15–20% year over year, and expects server DRAM contract prices to rise 13–18% quarter over quarter in Q3 2026. Note what that second figure is: a contract price, negotiated between suppliers and large buyers. It is not the price you pay at a retailer, and the gap between those two things is the subject of the next section.

Why your SSD did not follow

This is the part that surprises people, and it is the clearest thing in our data. Over the same window that DDR5 moved +10.6%, our SSD index moved +1.2%. Essentially flat. If the memory shortage were a single weather system, both categories would be wet.

The reason is that NAND flash and DRAM are different supply chains with different economics. HBM competes with DRAM for the same wafers; it does not compete with NAND. So the mechanism that squeezed DDR5 has no equivalent effect on the drive in your build.

That said, do not mistake “flat at retail” for “flat everywhere.” TrendForce projected on 3 July 2026 that NAND flash contract prices would rise 10–15% QoQ in Q3, alongside conventional DRAM at 13–18%. Our observed retail prices have not moved with that forecast — yet.

The same kit, two prices

Here is the finding with the most immediate cash value, and you will not get it from a market report. When we see the same SKU in stock at more than one retailer on the same day, the prices are frequently not close:

Identical part number. Same day. $4.99 apart. In a market moving this fast, retailer pricing desynchronises — each one is repricing its own inventory on its own clock, and for a stretch they simply disagree. If you are buying a quantity of these, the difference between checking one retailer and checking two is a real number on a real invoice.

What to do if you are buying now

We are not going to tell you to buy now or wait — we do not forecast, and anyone who tells you confidently where DDR5 lands in November is guessing with a straight face. What the data supports is narrower and more useful:

• Price per gigabyte, not per kit. A 64GB kit at $850 and a 32GB kit at $520 are not 1.6x apart, they are $13.28/GB against $16.25/GB. The bigger kit is the cheaper memory. Kit-level pricing hides this constantly.
• Check both retailers, every time. The spread above is not a one-off; it is what a desynchronised market looks like. Two tabs, thirty seconds.
• Re-quote right before the PO. A quote from three weeks ago is a historical document now. Whatever your approval cycle is, the price check belongs at the end of it, not the start.
• Do not read a DRAM move across to storage. They are separate markets on separate supply chains, and our index shows them behaving differently right now. Budget them separately.
• If you buy at volume, ask about an LTA. The protection large buyers get is contractual, not incidental. It is worth knowing whether the door is open to you.

How we measure this

You should be sceptical of price charts, so here is exactly what is behind ours.

• Real listings only. We scrape live product pages at Newegg and B&H, both of which permit it in robots.txt — we check, and we record the check. Amazon is API-only via Keepa and never scraped. Micro Center blocks robots.txt inspection outright, so we dropped it rather than work around it.
• Nothing is invented. Demo and seed rows are flagged in the database and blocked from public pages at the database layer, not by convention. An out-of-stock listing records a null price rather than carrying the last one forward, because carrying it forward would be inventing an observation that never happened.
• The index formula. Per SKU, take the lowest in-stock price across retailers — the best price actually obtainable. Divide by capacity. Take the median across SKUs, so one weird listing cannot drag the series. Normalise to a frozen base of 100.
• We withhold thin categories. A median needs at least 5 SKUs to mean anything. DDR4 currently sits under that floor in our coverage, so we do not publish a DDR4 median — you will see it named as withheld above rather than quietly omitted.
• Our baseline is short. Live tracking began in June 2026. We would rather show you three honest weeks than a year of backfilled guesses.

FAQ