For the last five years, the villain of the PC building world has always been the graphics card. We grew used to the headlines about "scalpers", crypto-miners, and the agonising wait for an NVIDIA RTX card to drop to a price that didn't require a second mortgage. The Graphics Processing Unit (GPU) was the loud, expensive diva of the industry, demanding all the attention and all the money. But while we were all watching the GPU market stabilise in early 2026, a different, quieter crisis was building in the background. Now, it has arrived, and it is hitting us where we least expect it: in the humble, boring stick of RAM.

We are currently stuck in a brutal memory shortage that has seen prices for standard DRAM skyrocket between 30% and 60% in just the last few months. In some extreme cases, spot market prices for specific DDR5 modules have nearly tripled. The era of cheap, abundant memory is over, and the reasons why reveal a fascinating and terrifying shift in how the global hardware supply chain now operates. To understand this, we have to look at the relationship between the GPU and the RAM stick and how the insatiable hunger of artificial intelligence has turned them from partners into rivals for survival.

The GPU parallel: a new kind of scalping

To understand just how bad this RAM crisis is, we need to compare it to the GPU nightmare of 2020-2021. Back then, the shortage was driven by a mix of pandemic supply chain breaks and the cryptocurrency boom. Miners were buying up every GPU they could find to crunch numbers for Ethereum, leaving gamers with nothing. The "scalper" became the enemy, a middleman artificially inflating prices.

The situation with RAM in 2026 is similar in effect but different in cause. There are no guys in vans buying up RAM sticks outside Best Buy. This time, the "scalping" is happening at the factory level, and the "miners" are the world's largest tech companies. The entities buying up the supply aren't individuals; they are hyperscalers like Microsoft, Google, and Meta, and they aren't buying consumer RAM sticks. They are buying the raw manufacturing capacity that would have been used to make them.

The steps that led us here are a masterclass in how a single technological pivot, the AI boom, can cannibalise an entire legacy industry.

Step 1: the vampire in the room (HBM)

The first and most critical step in this crisis is the physical reality of silicon manufacturing. Memory chips are made with silicon wafers, and there are only a limited number of factories (fabs) in the world that can produce them, primarily owned by three giants: Samsung, SK Hynix, and Micron.

In the past, these fabs churned out millions of standard DDR4 and DDR5 chips for our laptops and gaming PCs. But the rise of generative AI created a desperate need for a different kind of memory called High Bandwidth Memory (HBM). AI accelerators, like Nvidia’s Blackwell or H100 chips, cannot run on standard RAM; they need the massive speed of HBM to function.

Here is the problem: HBM is incredibly inefficient to make. It requires a die size that is significantly larger than standard DRAM. Industry reports indicate that HBM production consumes approximately three times the wafer capacity of standard DDR5. This means that for every one AI memory chip a factory produces, it sacrifices the ability to produce three chips for your home PC. It is a zero-sum game. Every wafer allocated to the "vampire" of AI is a wafer taken away from the consumer market.

Step 2: the producer pivot and the death of crucial

If you are a manufacturer like Micron or SK Hynix, this is not a crisis; it is a gold mine. HBM chips command profit margins that are 5 to 10 times higher than standard RAM. So, these companies have aggressively shifted their production lines. They are not just prioritising AI; they are abandoning the consumer.

The most shocking casualty of this shift occurred recently when Micron Technology announced it would effectively discontinue its consumer-facing brand, Crucial. For decades, Crucial was the go-to brand for affordable, reliable memory upgrades. Its exit is a massive signal that the big players no longer see value in selling $50 sticks of RAM to gamers when they can sell multimillion-dollar HBM contracts to OpenAI.

Other manufacturers are following suit, albeit less drastically. Samsung and SK Hynix have diverted over 30% of their total DRAM capacity solely to HBM production. They are betting the farm on the AI boom, leaving the traditional PC market to fight over the scraps. This "producer pivot" has created a structural shortage where supply is being artificially constricted to chase higher margins.

Step 3: the "Legacy" trap and programmed scarcity

The final step in this price spiral is the treatment of older technology. You might think, "Fine, DDR5 is expensive; I'll just stick with my older DDR4 system." But the industry is ahead of you. Manufacturers are engaging in what critics call "programmed scarcity" for legacy DDR4 memory.

Because DDR4 is older and cheaper, it has the lowest profit margins. In a world where wafer space is precious, it is the first thing to get cut. Manufacturers are shutting down DDR4 lines faster than the market is naturally transitioning away from them. This has created a paradoxical situation where the "old, cheap" technology is seeing faster price hikes than the new stuff because nobody wants to waste factory time making it.

The consumer reality: 64GB is the new luxury

What does this mean for the average user in 2026? It means the definition of a "budget" PC has fundamentally changed. For years, 16GB of RAM was the standard, and 32GB was a nice upgrade. Now, with prices for a decent 32GB kit jumping from $149 to over $239, and 128GB kits reaching dizzying heights of over $1,100, builders are being forced to downgrade.

The dream of "futureproofing" your machine with 64GB or 128GB of RAM is becoming a luxury reserved for the wealthy or the professional. We are seeing a return to the days where users must calculate every gigabyte, closing browser tabs and managing background processes because they simply cannot afford the hardware headroom we took for granted just two years ago.

Conclusion: no relief in sight

The grim reality is that this is not a temporary blip. Unlike the crypto crash, which happened relatively quickly, the AI infrastructure build-out is a long-term project. Industry forecasts suggest that this "supercycle" of high prices will persist well into 2026 and potentially through 2027. The new factories being built in the US and Europe won't be online and fully operational until 2028 at the earliest.

Until then, we are living in the shadow of the AI giant. The graphics card may have been the loud villain of the past, but the RAM stick has become the silent killer of the present budget. We are learning the hard way that in the digital age, memory is not infinite, and right now, the machines need it more than we do.