Future Proof Tech Briefing · July 2026
RAMmageddon Part 9: The Kitchen-Table Moment
A monolith, a kitchen table, and the question of who is actually in control.
2001
In 1968, Stanley Kubrick placed a black monolith in front of a band of proto-humans. They don’t understand what’s standing there. They touch it, they hesitate, they back away — and then that one artifact changes everything that comes after. The most famous cut in film history follows minutes later: a bone hurled into the air becomes a spaceship. A tool turns into the leap into a new age.
The monolith never explains itself. It just stands there — and forces everyone in front of it to make a decision: understand it, or get run over.
2026
The monolith now stands on your kitchen table. It’s nine centimeters tall, green, and its name is DDR5.
A look at the numbers: a 16 GB DDR5-5600 module cost around 45 euros in the summer of 2025. By the end of 2025 it was 60, and in 2026 a 32 GB kit runs about 300 euros. Year over year, individual DDR5 kits are up as much as 440 percent according to market data; DRAM chips in general, more than 170 percent. In the third quarter of 2026 alone, prices climb by up to another 18 percent. Gartner expects a combined DRAM-and-SSD increase of 130 percent by year’s end; Jefferies, another 40 to 50 percent in the third quarter and 30 to 40 percent in the fourth.
What was an abstract line in hyperscaler investment reports across the first eight parts of this series now sits as a price tag next to your laptop and your controller. Not in a data center “out there.” On your table. In your shopping cart. In the bill for the memory upgrade that cost a fraction of this a year ago.
That is the kitchen-table moment. The point where data-center economics stops being a topic for procurement teams and architects and starts being visible to anyone who owns a device. I call it the consumerization of visibility: it’s not the technology that’s being democratized — it’s the scarcity.
Why this is not a cycle
Past memory crises were waves — they came, they went. This one is structural. Manufacturers are shifting capacity into the lucrative high-end segments for AI servers. Vendors like Adata speak openly of an “acute supply shortage” and serve large cloud customers first, consumers last. Micron has signed long-term, non-cancelable supply contracts worth around 100 billion dollars — a contractual price floor that holds even when new fabs come online, because their capacity has long since been allocated to hyperscalers.
Even buyers with traditionally enormous purchasing power are feeling the squeeze. And in late June 2026 the issue even landed in court: a class action against Samsung, SK Hynix, and Micron over allegations of illegal price fixing. Justified or not — it shows how much the shortage is now perceived as a system beyond any individual’s control.
HAL
And this is exactly where a price story becomes a question of control.
Kubrick never explained the monolith. But he did show us HAL 9000 — the system that at some point starts deciding for itself. HAL wasn’t evil. HAL was what happens when we hand control of our own technology to a logic we no longer fully grasp.
In 2026, HAL is not a shipboard computer. HAL is the procurement logic of a handful of hyperscalers that effectively co-decides how much memory is left over for the rest of the world. The question is not whether the monolith is standing on your table — it already is. The question is whether you are Dave: calm, methodical, hand on the switch. Or whether you let it roll over you.
As an aside: HAL is just one letter away from IBM in the alphabet. Clarke swore it was a coincidence — and maybe that’s the real punch line. The systems that take control away from us don’t need a hidden message. They’re simply there, one letter over, doing their work.
The Survival Guide
At ISC 2026, Lenovo took exactly this stance. Martin Hiegl laid out what the company calls its “RAMageddon Survival Guide” — with the half-serious, half-smiling message that prices will “never again” be as low as last year. What that means: a new normal that settles in around 2030 at the earliest, permanently above 2024 levels.
The guide consists of five steps, and they are unspectacular — in the best sense: honestly review your requirements. Optimize operations. Choose the right CPU. Adapt the application. And, as the biggest lever: move memory-intensive workloads onto GPUs where it makes sense. AI inference doesn’t necessarily have to burden the host’s DDR5 — GPUs bring their own memory (HBM or GDDR6). Routing that load away from main memory lowers DDR5 demand and makes the budget a bit more independent of the price spiral.
The common thread is not panic stockpiling — it’s architecture. If you know today which workloads are truly memory-hungry and which can be moved, if you build your infrastructure to withstand scarcity instead of being at its mercy — you stand before the monolith like Dave, not like the proto-human.
This, incidentally, is the point where cloud-native stops being a buzzword. If you orchestrate workloads cleanly, keep them portable, and can steer them across clusters — across vendors if need be — you turn a hardware shortage into a pure planning exercise. That’s what cloud-native competence is about at its core: not the coolest technology, but the ability to stay sovereign when the fundamentals run short.
From a position of control
The monolith forces a decision. That was true in 1968, and it’s true in 2026.
You can wait until the price forces your hand. Or you plan now — from a position of control, not under pressure.
Jens Klasen is a tech evangelist at CID GmbH and writes the RAMmageddon series on the 2026 memory crisis and its consequences for enterprise IT architecture. You’ll find all previous installments in my newsletter and at klasen.ai.