Let's see how we can help you!
Leave a message and our dedicated advisor will contact you.
Send us a message
0/10000
Leave a message and our dedicated advisor will contact you.
If you planned to "just add some RAM" to breathe new life into your desktop in 2025, you likely hit a wall. A reinforced concrete one, at that. Memory, which for years was the cheapest and easiest way to speed up a machine, has suddenly become a luxury good.
A 64 GB kit rivalling the price of a game console? That is our new, painful reality.
It is easy to shout "It’s because of AI" and end the discussion. But reality is far more nuanced. It is a fascinating game of interconnected vessels, where silicon physics mixes with corporate calculation and global politics. Let's look at the five "horsemen of the memory apocalypse" that are rarely spoken about out loud.
Let's start with the fundamentals. In the world of semiconductors, you cannot cheat physics. We all know NVIDIA needs massive amounts of memory for its AI accelerators. But few realize the physical cost of this production.
HBM (High Bandwidth Memory), used in AI, isn't just standard chips laid flat. These are silicon skyscrapers—multi-layer stacks connected by vertical tunnels.
Key Takeaway: Producing the same capacity in HBM technology consumes about three times more silicon wafer surface area than standard RAM.
Put simply: every silicon wafer dedicated to AI is effectively three wafers that did not become memory for your PC. This is capacity cannibalization on an unprecedented scale. SK Hynix has sold out its capacity for years in advance, effectively removing a gigantic portion of supply from the consumer market. It’s not greed—it’s a lack of space on the production line.
In normal times, we would count on Samsung—the giant—to flood the market with product and stabilize prices. But 2025 is not normal. Samsung, struggling with efficient HBM production (and facing low margins around 30%), made a surprising pivot.
Instead of saving gamers and everyday users, the company redirected massive capacity (estimated at 80,000 wafers per month!) toward server RDIMM production.
Why? Pure business math. The AI frenzy created a gap in server rooms too, pushing margins on RDIMM modules to an astronomical 75%. Samsung engaged in strategic arbitrage: sacrificing the consumer market to make a fortune on corporations. It’s a decision that looks great in an Excel sheet, but terrible in individual customers' wallets.
This is perhaps the most counter-intuitive point on this list. Usually, "old" hardware gets cheaper. Yet in 2025, prices of aging DDR4 memory started chasing brand-new DDR5. How is that possible?
It is a classic supply shock.
The result? Supply evaporated, but demand remained. DDR4 prices, which scraped the bottom in March ($25), skyrocketed to $85 in just a few months. This is a brutal lesson for companies maintaining older infrastructure—delaying upgrades just became very expensive.
Hardware isn't the only culprit. A silent killer of low prices is a date: October 14, 2025. This marks the end of support for Windows 10.
For corporations, a lack of security patches is a risk they cannot accept. Instead of paying Microsoft a "ransom" for extended support (the price of which doubles every year), companies have gone on a massive shopping spree.
The global PC market has revived, but new "AI PC" class machines are no longer satisfied with 8 GB of RAM. The standard is becoming 16 or 32 GB. This forced hardware refresh cycle acts as a multiplier: more computers are being sold, and each needs several times more memory than its predecessor. It is pouring gasoline on the fire.
Finally, a factor we most often forget. A chip is essentially electricity enchanted into silicon. Semiconductor production is unimaginably energy-intensive.
In South Korea, industrial electricity prices have risen by 75% over four years. This cost must be passed on to the customer. Add to this the US-China trade war and tariffs raising import prices by another 20-50%, and we get a "hard price floor." Even if demand were to drop, production and logistics costs will not allow prices to return to the levels of years past.
We are dealing with a "perfect storm." The consumer market has been de facto forced to subsidize the Artificial Intelligence revolution. Every dollar we overpay for RAM today funds the arms race in AI and servers.
Experts have no illusions—the days of cheap memory are gone. Relief might only come around 2027, when new factories come online. I leave you with a question worth asking while looking at store prices:
Are we witnessing a moment where the PC has become "collateral damage" of the AI era, and high component prices are a new, permanent "tax on progress" that we will be paying for years?
Aleksander
RAM prices have skyrocketed due to a "Perfect Storm" of five factors:
The short answer is: not anytime soon. Analysts predict that high RAM prices will persist throughout 2025, with potential relief coming only in the first half of 2026. Waiting for a "price dip" is risky—if you need performance today, buy today.
Due to a market anomaly where DDR4 prices have caught up with newer DDR5, investing in the old platform no longer makes economic sense. It is only viable for minor repairs. If you are building a new PC or planning a major expansion (e.g., to 64 GB), DDR5 is the only cost-effective choice.
Requirements have shifted. In the era of "AI PC" and Windows 11, 16 GB of RAM is the absolute minimum for office work. For comfortable gaming, streaming, and creative workflows, 32 GB of RAM has become the new sweet spot.

Chief Technology Officer at SecurHub.pl
PhD candidate in neuroscience. Psychologist and IT expert specializing in cybersecurity.

The number of security incidents involving children has hit the roof - over 600,000 reports. We analyze NASK and Police data: from deadly TikTok challenges, through Roblox scams, to dramatic legal loopholes in sexting.
Can artificial intelligence experience trauma? We explore the fascinating and disturbing results of an experiment where algorithms underwent therapy sessions. The result? Synthetic psychopathology.
An analysis of 100 trillion tokens debunks myths about artificial intelligence. It turns out that instead of office productivity, roleplay reigns supreme, and users bond with models like in the Cinderella fairy tale.
Loading comments...