Memory and SSD prices are rising to levels that haven’t been seen in years. SysAdmins who purchased systems a year or two ago, with plans to upgrade their memory and storage in 2026, are seeing prices increase instead of decrease because of the 2026 memory and storage shortage, rendering those plans too costly to implement. New server prices only make sense if these servers are bought with the tiniest complement of memory and storage.
Why is this happening? What can the memory/SSD buyer do about it? Read on to learn what’s behind the changes and to understand how to deal with the issue.
Why are RAM and SSD prices spiking?
Everyone is fascinated with AI these days. To respond to this, major hyperscale datacenters are all investing heavily in AI computing systems.
How heavily?
The combined capital spending of Alibaba, Alphabet, Amazon, Apple, Azure and Meta totaled a cool $137 billion in the first quarter of 2026. As a comparison, these companies spent only $83 billion in the first quarter of 2025, and only $51 billion in 2024’s first quarter.
The chart below shows what that looks like graphically:
Notice how dramatically the ramp increases starting in 2024.
Now DRAM and NAND flash makers need to plan their production capacity well in advance, since it takes at least two years to bring a new wafer fabrication plant (a “Fab”) into volume production, and they have not been provided solid evidence that these hyperscalers would be investing in AI the way that they have. All this spending has led to a need for more fabs, and those fabs are being built, but they’re not yet in production.
Meanwhile, the hyperscalers have been increasing their spending ramp in a way that makes one quarter’s demand forecast obsolete the very next quarter. The planned new fabs just can’t catch up.
This has resulted in a NAND flash shortage that has driven enterprise storage costs to extreme levels, and it has similarly caused a memory price increase in the data center.
DRAM and NAND flash prices are rising, and that’s impacting system implementers in all businesses.
Supply constraints and AI demand driving costs
Why are prices increasing to the degree that they are? It’s because memory chips (both DRAM and NAND flash) are commodities.
A lot of people have trouble with the idea that something as technologically sophisticated as a DRAM or NAND flash chip could be a lowly commodity like rice, oil or aluminum. But a commodity is, in reality, anything that you can purchase from multiple sources. If you don’t care whose product you use, then price plays a dominant role in your purchase decision. You buy your parts from the cheapest supplier.
The sameness across all vendors’ products is driven by the JEDEC standards body. JEDEC brings OEMs and chip makers together, and everyone agrees on a specification for a chip that the vendors can all make, and the OEMs can all use. Anyone whose part matches the JEDEC specification can play in the market. OEMs avoid purchasing chips that don’t meet the JEDEC spec, for fear of becoming dependent on a single source for a part — this might put them into a “Line Down” situation where they can’t build their own product if their chipmaker has a problem (such as an earthquake, fire or strike).
Here’s a blow-by-blow of how this situation can make prices rise. Let’s say that there are three vendors for a part, and one of them has a lower price than the others. If demand spikes, then that vendor will find itself sold out, and the extra business goes to the other vendors. These other vendors see this as an opportunity and take a hard line during negotiations. The undersupplied OEM gives in, and accepts a higher price, rather than suffer from a line down. If any vendor increases its prices less than the high bidder, then its product will sell out too. It’s like falling dominoes.
Having learned that lesson, the vendors all raise their prices for their next sales to avoid selling out. If they still sell out, then they raise prices again. This continues until prices rise so high that their customers are either forced out of business or they pursue some different avenue. That other avenue might be to stop using DRAM size as a feature. So far this has only happened to a very small degree, so prices can continue to rise for the time being.
When will RAM and SSD prices drop?
This situation will stop at some point, but in this case it’s very difficult to tell when that will happen. Two factors can stop the price rise and will even cause prices to drop:
- Chip makers increase their production capacity until it catches up with demand.
- Demand falls, falling to a level that matches the chip makers’ capacity.
This is referred to as the “Supply/Demand Balance.”
Since hyperscaler capital spending (Capex) continues to be revised upward, it’s extremely unlikely that suppliers will be able to add capacity to match demand for anything within the next few years. But history has shown that it’s very possible that demand will slacken sometime in the next few years.
If we look back to the Internet bubble of the late 1990s, we see a very similar situation. When the Internet arrived, people recognized that it was a new technology that would change the world, much as AI promises to do today. Consider the many aspects of life that are dependent on the Internet’s connectivity, from driving directions to video conferencing to searchable web resources to watching television streams on demand. But the Internet’s capacity was overbuilt by 1999, resulting in a pause in investments in 2000 that led to a market collapse.
When that happened, some pundits looked to the past to find similar situations. One such event that happened in the US was an enormous railroad boom in the late 1860s and early 1870s. By 1873 the rail system was overbuilt, leading to a similar collapse. Of the country’s 364 railroad companies, 89 went bankrupt that year.
Objective Analysis expects for a similar cycle to occur in AI, where the high level of investments will lead to an overcapacity. Investors will slow their spending to the point at which demand falls below supply, and prices will collapse. When will this occur? There is no way of telling.
Timing your storage purchases
Where does that leave the data center professional trying to make the best purchase decisions? It’s not going to be an easy time.
One thing that would be unwise would be to purchase more than needed because of worries that resources won’t be available for purchase tomorrow. This can lead to excess resources whose value might depreciate at an alarming rate once the collapse begins. That devaluation will come straight out of the company’s bottom line.
A better approach would be to invest in storage management software that will help get the most out of the storage you already have. Some of these tools produce very good results simply by managing existing resources better than a previous method.
Deduplication is another good trick to make the best use of existing storage. This, once again, is a software-only approach that can yield surprising results.
Another option is very simple — implement an archiving policy, or tighten down on existing archiving policies. This can be done as a temporary measure, as the company is liable for retaining certain information for longer than necessary.
Strategies to offset rising SSD costs
Let’s say that, despite a diligent search for alternative ways to get by with existing storage, an organization finds itself unable to postpone storage upgrades. What are the options?
One would be to salvage older drives, securely erasing any sensitive data, and sell and replace them with larger drives. This might not find favor with security-minded bosses who have in the past destroyed storage devices when they decommission hardware, but good data destruction programs are available from a range of suppliers. In the end, this can help offset some of the upgrade costs, since today’s shortage has made more people open to purchasing used equipment. Meta has even published a paper about the way that it has salvaged older DDR4 RDIMMs (Registered DIMM) from decommissioned servers for use in newer DDR5-based servers.
The important fact, though, is to regard these upgrades as a series of incremental improvements, rather than as a single purchase made only once every few years. This sort of campaign involves more storage procurement paperwork, but it will ensure that an organization doesn’t spend far too much money by purchasing large amounts of storage immediately before a price collapse.
Rest assured that others are facing this issue as well – competitors are in the exact same shape, and the difference between an organization’s success and potential failure will stem from the amount of thought and effort expended to ensure that the absolute best use of storage is being made and that purchases are not being made ahead of needs at a time when prices are peaking.
Jim Handy is a semiconductor and SSD analyst at Objective Analysis in Los Gatos, Calif.
