In the fall of 2025, many people, myself included, opened their favorite hardware store to 'quickly grab another 32–64 GB of DDR5 for games, an IDE, and a couple of Docker containers'—only to close the tab in mild culture shock. The memory that cost a 'reasonable' amount in the summer suddenly cost almost as much as a mid-range graphics card.
In short, this isn't 'greedy stores' but the consequence of a rather complex restructuring of the entire DRAM market for AI servers and HBM memory. In this article, we'll explore what's happening at memory factories, why PC modules are suffering the most, what to expect in 2026, and how to make upgrade decisions if you're a gamer, developer, or just a hardware enthusiast.
What's happening in the RAM market right now
Let's start with the hard facts as of late 2025 to make it clear that the price increase is not a local anomaly in 'our' region.
Industry reports show a year-over-year increase in wholesale DRAM prices of more than 170% by the third quarter of 2025—one of the sharpest spikes in the history of the memory market.
Major manufacturers (including Samsung) have announced DRAM price hikes of 30–60% relative to September levels, with some categories (like automotive) seeing even higher increases.
According to TrendForce, in the fourth quarter of 2025, 'regular' DRAM (PC and mobile chips, excluding HBM) is getting another 8–13% more expensive quarter-over-quarter, and if you include HBM, the increase is up to 13–18%.
The retail picture confirms this: popular 32-gigabyte DDR5 kits, the 'sweet spot' for gaming (e.g., DDR5-6000 CL30), have seen their prices in euros increase by about 80–100% over two autumn months.
At the same time, market reviews note that by the end of 2025, the price gap between DDR4 and DDR5 has almost disappeared: following a reduction in the production of older DDR4 chips, their prices have risen so much that they are now on par with the more modern DDR5. This is particularly painful for those upgrading older platforms.
And here's the main point: the root cause is almost always the same—an explosive growth in demand for memory from AI clusters and data centers, which are consuming more and more silicon in the form of HBM stacks and server-grade DDR5.
How money is made in the DRAM industry (and why it matters)
To understand why any demand shock hits retail RAM prices so quickly, we need to take a brief look inside the economics of a DRAM factory.
The market is effectively an oligopoly.The bulk of DRAM is produced by a few companies. Meanwhile, the capex for building and upgrading factories is in the tens of billions of dollars, with a payback period of years. This makes memory production highly inertial.
Memory is a commodity product with cycles.Manufacturers plan production 'by the bit': how many terabits of DRAM will be produced per quarter. When the bit supply grows faster than demand, a price war begins and margins fall, after which some capacity is cut—and the cycle reverses.
A one or two percent imbalance breaks the market.Analysts note that even a 1–3% supply-demand imbalance in DRAM is enough to trigger a sharp price increase, which is what we're seeing now amid the prioritization of AI workloads.
And the most unpleasant part: as soon as manufacturers have the opportunity to get a much higher margin on 'AI memory,' they will shift capacity there, even if it hurts the prices of mass-market modules for PCs and smartphones. For them, this is rational behavior.
Where the silicon is going: HBM, servers, and AI
The star of the last two years is HBM (High Bandwidth Memory), a multi-layered memory soldered next to a GPU/accelerator on a single interposer. It's what feeds the training and inference farms for LLMs (Large Language Models) that are so popular in press releases.
Key points:
HBM is incredibly complex to manufacture.Multi-layered stacks, TSV contacts, and tight integration with the GPU—all of this reduces the yield of good dies and increases the cost compared to 'regular' DRAM.
HBM and classic DRAM compete for the same production lines.The transition to more advanced process nodes (1β, 1γ, and beyond) and high-margin HBM means that some of the capacity that could have been used to produce DDR4/DDR5 for PCs is being diverted to HBM and server memory production.
AI clusters consume more than just HBM.Surrounding each set of GPUs is a whole menagerie: server-grade DDR5/LPDDR5, buffer memory for network cards and drives, and caching solutions. This creates a 'secondary wave' of demand for regular DRAM, increasing the load on factories.
As a result, priorities are set as follows: first HBM and servers for large clouds and hyperscalers, then the mobile market, and only then consumer-grade desktops/laptops. PC enthusiasts find themselves at the very end of the line.
Why DDR4 and mass-market DDR5 were hit particularly hard
If you're running something like a Ryzen 3000/5000 or an Intel CPU up to the 12th generation and looking at DDR4 prices, the increase seems even wilder than for DDR5—and there are objective reasons for this.
Manufacturers are deliberately 'phasing out' DDR4.By 2024–2025, many were planning to gradually wind down DDR4 production, freeing up lines for DDR5 and HBM. At the start of 2025, this seemed logical: DDR4 was cheap, and demand was shifting towards DDR5.
Then the next wave of the AI boom arrived.When it became clear that margins on HBM and server-grade DDR5 were significantly higher than on budget DDR4 for desktops, manufacturers didn't restore previous DDR4 production volumes. As a result, the surplus quickly turned into a shortage. Analytical reports describe a nearly threefold increase in contract prices for DDR4 in just a few months.
DDR5 prices are rising more gently, but noticeably.According to several research groups, DDR5 prices rose more steadily and modestly than DDR4 prices in 2025, but in the consumer retail segment, kits still became significantly more expensive over the autumn.
In practice, this has led to a paradox: upgrading old DDR4 platforms has become almost as expensive as building a new system on a fresh DDR5 platform. This is especially noticeable in 'built-from-what's-available' systems and on the second-hand market, where prices for 'fancy' 32–64 GB DDR4 kits have skyrocketed to levels where they clearly don't belong.
What this means for gamers and developers in 2025–2026
Let's move from analysis to practical application: how to make decisions about RAM upgrades when the market is behaving so erratically.
1. When to buy—now or later
Public forecasts show that major analytical agencies expect DRAM prices to continue rising at least until the first half of 2026, with increases in the tens of percent depending on the segment.
This leads to several practical conclusions:
If the upgrade is needed for work (IDE + Docker + VMs, test clusters, etc.) and you're already regularly hitting swap, there's little point in waiting months for a hypothetical price drop in the current cycle. The probability that memory will become significantly cheaper in the coming quarters is quite low.
If the task is purely for gaming and you currently have a solid 32 GB of DDR4/DDR5, a reasonable strategy is to wait at least until prices stabilize or for the next wave of platforms (new generations of CPUs/chipsets) to upgrade everything at once.
It's important to understand that the current growth cycle is structural (related to the restructuring for AI) and not just a short-term shortage caused by a single incident.
2. How much RAM is really needed right now
Given the prices, it makes sense to soberly reassess your volume needs:
Gaming + light multitasking: 32 GB of DDR4/DDR5 is still the sweet spot. For most games plus a few browser tabs and an IDE, this is more than enough.
Development, multiple IDEs, Docker/WSL, light ML/analytics: 48–64 GB is a comfortable range, especially if you're running several containers, a database, and a browser with dozens of tabs simultaneously.
Heavy self-hosted stack, local LLMs, rendering: 96–128 GB and above. This is where the price increase hurts the most, but skimping on volume is a strange idea: you'll just hit swap and pay with your time instead.
A good tactic is to run through your typical scenarios while monitoring RAM usage (in Windows, Linux, macOS) and see what happens at peak usage over a week. This will help you understand if the upgrade is truly justified or if it's just a case of 'I want it because it looks cool'.
3. What to choose at current prices
A couple of practical tips to avoid overpaying in the current turbulence:
Buy a kit, not single sticks.Given the price increases and potential differences in chips and revisions, adding a single stick to an old kit is an unnecessary risk for compatibility and overclocking. It's better to sell the old kit and buy a new 2×16 or 2×32 kit.
Stick to the 'golden mean' for frequencies.For gaming and work desktops, a reasonable compromise right now is around 5600–6000 MT/s for DDR5 and 3200–3600 MT/s for DDR4. Upgrading to something like DDR5-7200 for a couple of extra FPS when prices have doubled is a questionable investment.
Don't overpay for RGB and exotic features.In a shortage, it's the 'fancy' product lines and rare capacities that get hit the hardest. If functionality is more important than the showcase look of your build, it makes sense to choose simpler models.
Be cautious with the second-hand market.Amid rising prices, the used memory market has become more active, but you need to be careful about the origin of the kits (mining, server use, aggressive overclocking) and not pay new-product prices for questionable listings.
4. Upgrade the platform for DDR5 or stick with DDR4
With prices for DDR4 and DDR5 leveling out by the end of 2025, the choice has become less obvious.
The general logic is as follows:
If you have a recent DDR4 platform stack (like a Ryzen 5000 or 10th/11th gen Intel) and already have 32 GB of decent memory, it's better to wait for your next platform upgrade unless you have a critical need, rather than rushing to DDR5 now 'just because it's trendy'.
If, however, you were already planning a CPU/GPU upgrade, then moving to DDR5 now is logical. You're unlikely to save much money by trying to expand a DDR4 configuration, but in terms of future prospects (support for new CPUs, higher base bandwidth), DDR5 is already the clear winner.
Essentially, the market itself has pushed many towards a logical decision: not to pour extra money into an old platform, but to plan a comprehensive upgrade when it becomes truly necessary.
What's next and how to approach it
According to current forecasts, strong demand for HBM and server memory will persist at least until mid-2026: major vendors and cloud providers are continuing to build AI farms, while DRAM manufacturers are just preparing new capacity and more advanced process nodes. This means you shouldn't bet on a quick 'rollback' price drop for memory like in previous years.
However, in the medium term (after 2026), a classic scenario is likely: when new factories and process nodes are up and running and demand stabilizes, the bit supply will finally overtake demand, and we'll once again see a cycle of relatively 'cheap memory'—but at new baseline speeds and with much higher specific consumption from AI workloads.
For us, as gamers and IT professionals, the main conclusion is simple but practical: treat RAM as a strategic resource, not a 'dirt-cheap consumable.' Plan upgrades consciously based on your real-world scenarios, don't fall for the hype, and don't try to 'outsmart the market' against a trend that is now being set by completely different budgets and players.