On August 20 I opened TrendForce's public spot price table and read one line twice. A 16-gigabit DDR4 chip, which is two gigabytes of memory, was averaging $90.59 that session. The same-density DDR5 chip, the newer standard, was averaging $53.77. The old part costs about 1.7 times the new one.
That inversion is not new. Counterpoint Research saw it last November, with DDR4 at $2.10 per gigabit against DDR5 at $1.50, a ratio of about 1.4. Counterpoint and TrendForce are different price trackers, so take those two ratios as direction rather than precision. The direction is that the gap has widened.
Now the second number. That same DDR4 chip appears on TrendForce's contract price table too, where the most recent figure available without a paid membership is $42.00, from late June. Spot was $90.59 on August 20. Those are two different markets, sampled seven weeks apart. The gap is real and I come back to it, but it does not explain a number more than twice as large.
Part of the cooling you are reading about is other people leaving the shop. That is demand destruction, the unlovely term for prices climbing until enough buyers give up that the climbing slows. It is not supply coming back. TrendForce expects the DRAM market to stay "extremely tight" through the third quarter.
Neither of those markets is yours, incidentally. You buy at retail, a third price, and TrendForce expects retail notebook prices to "rise across the board" as the expensive components work through inventory. That is the pass-through you are waiting on, and it arrives late and stays.
Tom's Hardware and technology.org got the thesis right and missed the base effect
Here is what TrendForce actually published on July 3. Conventional DRAM, the main memory in your machine, up 13 to 18 percent this quarter. NAND flash, the storage in your SSD, up 10 to 15 percent. Those are contract prices, the rates big buyers negotiate quarterly, and the two outlets I checked both said so and both got the reason right. Tom's Hardware wrote on July 4 that the cooldown came from consumer electronics manufacturers' "unwillingness and inability to absorb higher memory costs," and not from "improved supply." technology.org called contract prices "the ones large buyers actually pay," noted they were decelerating "even as spot climbs," and concluded that the slowdown was "a demand-side signal, not a sign the shortage is easing." They had the thesis before I did.
What neither of them has, in the versions I read on August 20, is the second cause, and TrendForce puts it in its own headline: gains moderate as consumer demand weakens "and High Base Effects Take Hold." A percentage is measured against the quarter before it, and the quarters before this one were enormous. That is not a footnote. It is half the explanation.
One number did travel badly. Tom's Hardware summed up the previous quarter as "roughly 60% jumps," a plural that reaches across the NAND clause sitting beside it. DRAM did run 58 to 63 percent. NAND ran 70 to 75. On July 24 technology.org repeated "roughly 60% in Q2," and its own closing source list cites that Tom's Hardware article by title and link, so the phrase has propagated.
The three quarters side by side make the shape plain.
Nothing here is a price cut. Every bar stacks on the one before it. The third-quarter bar stands about one-sixth as tall as the first, but it adds roughly half as many dollars, because its 15 percent is charged against a price the earlier forecasts had already roughly tripled. A smaller percentage of a much bigger number is not a small number. Treat all six as forecasts rather than history, too, because TrendForce moves them: it lifted its own first-quarter DRAM forecast from 55 to 60 percent up to 90 to 95 percent while the quarter was still running.
Three limits on my own numbers, since I have been strict with everyone else's. On whether spot itself is cooling I can see only one day, because TrendForce gates its 30-day price history behind a membership and my request came back refused. That session showed DDR4 up 0.54 percent, DDR5 up 0.94 percent, DDR3 the only line falling, and the DDR4 line swinging from $116.00 down to $42.00 within the session, its low matching the June contract price to the cent. The other limit is that contract price. TrendForce's 13 to 18 percent covers a full quarter, so growing $42.00 by the whole range gives about $49, which is a September figure and not a July one. July sits somewhere between the two. Either way, spot is running at roughly 1.8 to 2.2 times contract. The third limit is Gartner, whose newsroom refused my requests the way TrendForce's endpoint did, so both Gartner figures here come from Internet Archive copies of its own press releases, the memory forecast of February 26 and the semiconductor forecast of April 8.
One figure from that same Counterpoint note has traveled worse than the rest. It concerns HBM3e, the stacked, top-of-the-line memory that AI accelerators are built around. Counterpoint reported DDR4 at $2.10 per gigabit, "higher than that of even advanced HBM3e, which is hovering around $1.70," and it put all three numbers in a paragraph about spot prices. On July 24 technology.org reported that HBM3e "contracts at around $1.70 per gigabit." That verb is not Counterpoint's. It quietly converts a comparison of spot against spot into one of spot against contract, which is the precise confusion this piece exists to sort out, and it ran the eight-month-old figure undated and in the present tense.
The underlying number is stale rather than wrong. DDR4 was $2.10 per gigabit last November and my figure works out to $5.66 today, with the same tracker caveat as before. The commodity chip in a cheap laptop stick has nearly tripled since last autumn. I could find no current HBM3e price to set beside it, so I cannot tell you whether the inversion still holds. Judging by the direction of that DDR4 number alone, it probably does.
Phone shipments are forecast to fall 13.9 percent this year
Here is what buyers walking away costs, and it is where the demand destruction stops being an abstraction. IDC expects global PC shipments to fall 11.3 percent this year, with the fourth quarter down 20 percent. It expects smartphone shipments to fall 13.9 percent to 1.09 billion units, which it says would mark the steepest annual contraction in smartphone history. The year is not over, so that remains a forecast. IDC's full-year forecast also has the average phone selling for $550, up $100 on last year.
So, the verdict. Do not read "cooling" as a signal to wait for a dip. Gartner does not expect meaningful price relief until late 2027, and IDC does not expect it before the end of 2027. If you were going to add memory to a machine you own, that window closed months ago. If you are buying, the practical question is the one I now ask of every laptop: does it have real memory slots, or is the RAM soldered to the board for the life of the device? A machine you can open is worth paying for this year in a way it has not been for a decade.
Gartner expects the sub-$500 entry-level PC segment to disappear by 2028. That is a February call, and I could find nothing Gartner has published since that revisits it. Whatever replaces it will come with a cloud storage subscription, which is one way to sell you back the flash they left out of the box.
