If you’ve priced a new PC build recently, you already know the figures aren’t pretty. Less obvious from any single glance at a storefront is that this doesn’t represent a simple hiccup in the economics of the market. This represents a fundamental change in the approach to selling memory products by a certain number of major firms within the industry, and they have binding contracts which extend significantly beyond the lifetime of this particular problem.
Below we’ll outline what the figures actually say, why asking customers to hold off on purchases until the next cycle is around the corner might not be the best idea this time, and what should be done about the situation in the meantime.
The damage in dollars
Prices of memory and storage products of all kinds have climbed steadily since last September and have not shown any signs of slowing down, with RAM being disproportionally more expensive than SSDs, and prices climbing exponentially with capacity.
| Component | Sept. 2025 avg. | Now | Change |
|---|---|---|---|
| 32GB DDR5-6000 CL30 kit | $122.50 | $567.50 | +363% |
| 64GB DDR5 kit | $240 | $1,300–1,400 | +~483% |
| 32GB DDR4 kit | $62 | $243 | +292% |
| 2TB SATA SSD | $112.33 | $318 | +183% |
| 2TB NVMe SSD | $143.25 | $340 | +137% |
| 16TB hard drive | $350 | $800 | +129% |
Averages drawn from sampled product listings with tracked price histories on PCPartPicker.
Enterprise-scale kits suffer worse. A 128Gb DDR5-6400 ECC registered kit which once costed roughly $1,600 two years ago is functionally unavailable at that spec today; a comparable third-party 128GB kit now runs upward of $6,800, a 512Gb kit listed around $4,200 has been spotted for nearly $24,000 previously. A 4TB Samsung 990 Pro that sold for $390 in 2024 is now around $1,100 — a 254% increase its own.
Spot prices, the day-to-day wholesale rate, not which OEMs actually pay under contract but a good proxy for raw demand, tell an even more dramatic story. Since last July 16Gb DDR5 is up roughly 800%, 512Gb NAND wafer pricing is up around 678%, and 16Gb DDR4 — a technology being phased out not expanded — is up nearly 958%. One way to put it is that DDR4 spot pricing has been proceeding like a small-cap altcoin.
The real reason is long-term agreements, not a typical shortage
Each memory price spike of the past couple decades has eventually broken. Manufacturers overbought, demand dwindled and prices crashed back down, a boom-bust cycle so reliable that PC builders have learned to just wait it out. The industry itself is now trying to end the cycle deliberately, and the mechanism is the long-term agreement (LTA).
An LTA locks a customer’s future memory supply — typically about 50-70% of a manufacturer’s total — into a contract running three to five years, sometimes more. Instead of negotiating quarter to quarter, the largest five to sixteen customers of each supplier now pre-purchase years of capacity in advance. Here’s what that looks like at each of the major manufacturers, per their own earnings calls:
Micron’s roughly 16 LTAs already cover a fifth of its DRAM and a third of its NAND output, with the company aiming for half or more of total revenue under similar contracts once complete.MICRON — Q2/Q3 FY26 earnings call
The customers writing these contracts are the hyperscale cloud providers building out AI infrastructure, and they’re not shopping around anymore. They’re buying multi-year certainty.
Kioxia has said it’s tracking toward 50% LTA coverage by calendar 2028. None of this capacity is going to individual consumers, or even to the companies that sell directly to consumers — with the partial exception of Nvidia, whose consumer-facing business is now under 8% of total revenue.
An analyst quoted by the Korea Herald put the likely outcome plainly: LTAs probably won’t eliminate the memory cycle entirely, but should reduce earnings volatility for suppliers by locking in demand — how well this benefits consumers will depend on how well the contracts hold during the next downturn.
LTAs are transforming the industry from a three-to-five-year boom-bust cycle into a long-term, order-based model with pre-secured demand.— Choson Daily, on the shift in supplier strategy
For anyone hoping prices simply crashed the way they did after previous booms, it is a meaningfully more cautious forecast than the cycle we are used to.
What this means in practice
The value of the time-honored strategy of waiting for prices to collapse is diminished, as between 50–70% of production is committed years in advance, leaving less room for oversupply than one might expect, thus limiting downward pressure on prices.
The downside risk is lower, not higher, as the industry moves to a new normal. Analysts who have studied previous cycles believe that this time, the price fluctuation range will be narrower, with prices finding a higher floor, not lower, as compared to previous troughs.
The next catalyst for lowering prices is likely to be an economic shock. The only credible scenario for memory prices to fall back to previous lows involves a significant “AI bubble,” which would have far-reaching implications beyond just your component budget.
Why this is good news for cloud providers
There is one obvious winner in this scenario, which is not difficult to identify if one knows where to look: as on-premises hardware upgrades become prohibitively expensive, hyperscalers become even more desirable upgrade partners.
AWS announced in its Q1 FY26 earnings briefing that the price and supply dynamics in memory markets represented a “tailwind” to its cloud business, enabling enterprises to migrate to AWS from other on-premises IT solutions and helping to diversify its enterprise customer base.
This is reflected in the revenue line: Google Cloud’s revenue grew 82% year-over-year in its most recent set of earnings, while Microsoft’s Azure and other cloud services revenue climbed 43% – and AWS’s cloud revenue grew 36.7% – the fastest increase in eighteen months.
According to Trendforce estimates, combined spending on DRAM and NAND will comprise 47% of total CAPEX for the nine major cloud service providers in 2026 – 68% in 2027. As the overall cloud CSP CAPEX is estimated to rise to $922 billion to $1.38 trillion, the dynamics impacting your NAND purchases will also affect the capital expenditures of the world’s largest technology companies.
The politics layer nobody’s pricing in
Memory has become, in effect, inextricably entangled with trade policy, which is a novel development for the history of personal computing. Two Chinese manufacturers of memory chips, CXMT and YMTC, have seen their rise to prominence accelerate to the point where they’ve begun to challenge the established “Big Three” NAND producers, and one of them has already joined their ranks.
Meanwhile, another Chinese manufacturer has raised its share of the global DRAM market by revenue from 4% to 10% year-over-year, according to Counterpoint Research.
Both of these trends have had clear political consequences, as the recent letter from Senator Chuck Schumer to Apple demanding that the latter cease and desist from purchasing memory from Chinese firms demonstrated.
Commerce Secretary Howard Lutnik has also publicly stated that he did not consider Chinese memory to be a desirable option for U.S.-based firms, despite acknowledging that the industry as a whole needed an alternative.
The problem with that assessment, of course, is that the U.S. is currently responsible for the production of less than 3% of global DRAM, with Micron’s CEO himself estimating that the newly announced plants in New York and Idaho would only raise that figure to around 10% by 2035, with a significant proportion of that allocated to HBM and enterprise-class RAM.
Both Schumer and Lutnik have proposed “solutions” that, if anything, exacerbate the problem, with the former floating the idea of linking the allocation of CHIPS Act subsidies to the expansion of customer bases and the latter suggesting the use of the Korean War-era Defense Production Act to appropriate a portion of memory production for government use.
However, even if none of these theoretical options appeal to you, it is important to understand that the situation in itself is perfectly normal: attempts to exclude a particular source of supply usually yield exactly one result, namely, reduced supply from other sources.
Is there any good news?
NAND and DRAM are not the same thing, and it is worth looking at the differences. TrendForce, one of the respected memory market research companies, estimates that NAND production will cross into positive territory in terms of supply-demand ratio sometime in 2027.
The appearance of new production layers based on more advanced technologies will contribute to this trend, while DRAM production is expected to do the opposite in the same period. In part, this is already reflected in the prices of some memory manufacturers.
Take the example of Silincon Motion, whose president stated that the retail SSD market has all but disappeared as a separate category, and shortages were expected to intensify in 2023 before NAND prices began to decline.
Therefore, we can expect a decrease in NAND prices in 2027, assuming the projected production figures are accurate.
However, this forecast does not extend to DRAM, and the reason for this is simple: the same wafer goes into servers and HBM production for AI accelerators and graphics cards, and is not available for consumer products, at least until 2035.
What can be done about it
First, buy what you need and not need right now. The headroom that you will not use for two years is the most expensive memory in terms of 2026 prices. Second, reuse the memory and storage you have at hand if possible.
Third, avoid buying DDR4 memory if at all possible, as manufacturers are quickly ending their support for this type, and its share price was the most volatile among all the options listed above.
Fourth, watch out for NAND more closely than for DRAM. If and when the supply-demand figures change for the better, which is expected to happen around 2027, this will be reflected in the price of SSDs. By contrast, there is no light at the end of the tunnel for DRAM.
Finally, keep in mind that prices change rapidly, and manufacturers do not announce them for longer periods. Therefore, it makes sense to make a purchase decision as close to the date of acquisition as possible, and not to rely on information more than a few weeks old.
The situation with computer memory in 2023 is textbook textbook market manipulation in the purest sense of the term. The largest memory manufacturers have begun to reorganize their work to prioritize long-term contracts over short-term sales volumes.
This allows them to optimize costs and cash flows, something that is especially important to them given the high capital intensity of the production of memory chips.
This strategy has nothing to do with any particular conspiracy, but rather reflects the direct and public statements of these companies’ executives on the topic. In practice, this means that the cost of any random piece of memory will tend to be higher, its availability lower, and its volatility more difficult to predict.
This article contains reporting and analysis from Gamers Nexus, owner of this copyright.