For the past few months memory prices have been skyrocketing. This has been fueled by an interesting combination of extraordinary AI spending and HBM’s low gigabyte production per DRAM wafer. Given these facts, which I will explain in detail in this post, just how high will memory prices go? And, after that, how long will prices stay this high? The Memory Guy has been getting asked that question a lot lately, so I will explain today’s DRAM prices and their behavior in this post.
A High-Tech Commodity
It is difficult for many folks to come to grips with the fact that most memory chips are commodities. A commodity is not necessarily something cheap, in fact gold and platinum are commodities and are traded on commodities exchanges, along with steel and aluminum. Commodities don’t have to be low-tech either.
If you want to know if something is a commodity, all you have to do is ask: “Can you buy a satisfactory replacement from multiple suppliers?” If you can, then price plays an enormously important role in the marketplace, since a part’s price is the leading reason for a buyer to choose one supplier over another. This is what makes a product a commodity.
Memories are pin-compatible with one another. You can replace a certain SK hynix DRAM chip with an equivalent one from Micron or Samsung. The same is true of NAND flash. Most OEMs can make a switch like this almost immediately. When there’s a shortage the ability to switch allows them to change suppliers if their preferred supplier runs short. When supply is abundant, they can use their ability to switch suppliers to force these suppliers to bid their lowest price for a share of the business. This causes prices to be extremely volatile — when there’s a shortage they increase, and when there’s an oversupply they go down.
Of course, suppliers try to increase production when there’s a shortage, and sometimes close facilities when there’s an oversupply, but there’s always a lag between the decision and when production volume changes, so shortages and oversupplies cannot be closely managed, and this extends the duration of the imbalance.
The natural effect of this is that the market cycles back and forth between shortages and oversupplies, and prices go up and down as a result. This has been the case for the life of the semiconductor market.
In economics, this is referred to as the Commodity Cycle.
All commodities undergo such cycles, whether agricultural, mineral, or even services like shipping and airlines.
What Limits Price Rises and Falls?
Given this dynamic, what limits the heights and depths of memory price moves?
Today it appears that there is no limit to price increases. DRAM spot market prices, which always behave much more dramatically than the contract prices that account for over 95% of total market revenues, have increased to over four times their level last September, and NAND flash spot prices are up to almost seven times their September levels. Contract prices, while more subdued, have still risen to more than double their 2025 levels.
Although consumer products are regulated in the US to disallow price gouging, that is not the case for semiconductors. But at a certain point market forces start to take effect.
In the case of memories that point begins where the price of the memory chip rises so high that it kills off the market for the end product. We are not there yet, since memories account for only a fraction of the cost of a cell phone, PC, or server, thereby limiting a system’s end-price impact from a 4X price rise in its memory components. Other factors that I won’t detail here also limit the impact of memory price increases on PC and cell phone demand.
As for a price fall, that is limited to a chip’s production cost, since selling chips below cost leads to government trade sanctions. But that’s not today’s concern.
How Did AI and HBM Lead to This?
Probably everyone reading this post has seen a chart like the one below, illustrating how astronomical hyperscale datacenter capital spending has become.
This rendition shows combined quarterly capital spending for six leading hyperscalers. It was a part of my presentation in February’s annual Chiplet Summit, where I was invited to give the opening plenary. (Electronic Design Magazine has published a video of that presentation.) During that talk I explained that quarterly spending for the six companies in the chart had reached $70 billion by the third quarter of 2024. By the fourth quarter of 2025 that number had nearly doubled to just below $130 billion.
The difference between that and the $30-50 billion that was the prior spending level was all going to AI systems, most of which use GPUs, TPUs or some equivalent. All of these GPUs and TPUs use internal HBM to provide an impressive amount of DRAM immediately beside the processor to provide the best possible bandwidth at the lowest possible latency.
Even if these systems used standard DDR DRAM there would have been a shortage as hyperscale datacenters, the application that already consumed the largest share of DRAM, tripled its quarterly capital spending in only two years. That’s why NAND flash is also having a shortage. But the shortage is worse for DRAM, and that’s because the DRAM that these systems use is largely HBM.
HBM is not standard DRAM. Although it is made using the same DRAM process as DDR, HBM requires a special chip design that only produces one third as many gigabytes on a DRAM wafer as the same wafer would produce of DDR. Even if there had been no AI spending binge, if the DRAM business had shifted from 0% HBM to 10% HBM in a quarter or two supply would not have kept pace with demand.
This just works to make today’s market shortage a lot worse than it would have been if these processors could use standard DDR.
Wrap-Up
To summarize, prices aren’t as high as they could be, and there’s the possibility that they will raise even higher if today’s overheated AI demand doesn’t suddenly cool down. Although DRAM makers are adding capacity at a speedy pace, it’s nowhere near enough to bring the market into balance anytime soon.
My company, Objective Analysis, focuses on situations like this, and we guide our clients into optimizing their decisions in whatever situation the market presents. If you would like for your company to have a successful edge over its competition, please contact us to explore ways that we can work together to your firm’s benefit.