Future Proof Tech Briefing · August 2026
RAMmageddon, Part 10: The Spice Is Allocated
A desert planet, a stillsuit, and the question of whose wasteland catches up with you in the end.
1965
In “Dune,” Frank Herbert builds an economy around a single substance. Spice cannot be manufactured, only harvested. It comes from a single planet. And whoever controls it controls everything built on top of it — space travel, trade, politics.
What’s interesting about Herbert’s construction isn’t the scarcity. It’s the allocation. It isn’t price that decides who gets Spice, but the Spacing Guild and its contracts. If you’re not in the allocation, you can have all the money in the world — you’re not flying.
2027
July and August are traditionally when memory quotas for the following year get negotiated. This year the negotiations were over unusually early. According to a DigiTimes report citing industry sources, the three big manufacturers have already fully allocated their DRAM and HBM capacity for 2027. There is no confirmation from the manufacturers, and the report should be read accordingly. But it fits a picture that hasn’t shifted in months.
The form is what’s remarkable. Long-term contracts running three to five years. Prepayments even where no long-term contract materializes. And allocation quotas that, according to the report, frequently land at 60 to 70 percent of the requested volume.
In Part 9, the monolith lay on your kitchen table as a price tag. That was the visible phase. Now comes the invisible one: prices you still negotiate. Allocation you simply receive.
I call this the allocation economy. The point where procurement stops being a commercial discipline and starts being an architectural one. Because no better buyer helps against a quota. The only thing that helps against a quota is needing less.
The Guild
Who gets served first is no secret: cloud providers and large AI companies. Then everyone else. PC and smartphone manufacturers are expected to have noticeably less left over in 2027 than in 2026 — and standing in that same line is the mid-sized company that wants to procure its own hardware.
That upends a logic that held for twenty years. Your own infrastructure was the predictable option: invest once, depreciate, plan. The cloud was flexible but expensive on an ongoing basis. That calculation looks different when your own hardware becomes not just more expensive but uncertain to get at all. A quote with a price is a different thing from a quote with a delivery date.
And the hyperscalers sit on the right side of the allocation. They buy earlier and in volumes no systems integrator can match. Anyone who needs capacity in 2027 is more likely to get it there — on terms that will likewise reflect the scarcity. The Guild flies. It just doesn’t fly for free.
The wrong question
This is where the discussion goes wrong in a lot of organizations. It gets framed as a directional decision: private cloud or hyperscaler. On premises or public. As if you had to commit now.
You don’t. What you have to do is prepare to be able to commit later.
Nobody can say with any confidence today what the capacity situation looks like in 2028. The fabs meant to relieve the pressure are still being built. Anyone moving irreversibly in one direction today is betting on a scenario. Anyone keeping the decision open retains freedom of action — and under scarcity, that is the genuinely scarce good.
So the question isn’t: where are we going? It’s: will we still be in a position to decide that ourselves in 2027 — or will availability decide for us?
The stillsuit
Herbert’s Fremen didn’t survive on Arrakis because they had more water. They survived because they invented the stillsuit: a suit that reclaims every drop of moisture the body gives off. Loss per day: a thimbleful.
That’s the answer to an allocation economy. Not getting more. Losing less.
And here is the difference between Arrakis and what we’re talking about. Herbert’s desert was found. It was there before the first human set foot on the planet. The wasteland that concerns you is self-made.
It grew over years because memory was cheap. As long as water costs nothing, nobody builds a stillsuit. Oversized systems with statically assigned memory, because nobody ever measured. Monoliths that have to be scaled as a whole even though only one part is under load. The same data in triplicate, because three systems each maintain their own copy. Applications that have been running for eight years because nobody is allowed to switch them off and nobody remembers who actually needs them.
That’s the DigitalWasteland: not the market environment, but the sprawl inside your own house. It doesn’t come from decisions — it comes from their absence. Do nothing and you end up there inevitably; not because something goes wrong, but because nothing gets cleaned up.
Until now that was a question of hygiene. Now it’s a question of procurement. Every gigabyte an application doesn’t need is one you neither have to be allocated nor have to rent. In a market with stable prices, efficiency was an optimization. In a market with allocation, it’s strategy.
And here lies the point that got lost in the modernization debate of recent years. Application modernization was sold as the path to faster development, better maintainability, modern architecture. All true, all hard to quantify — which is why such projects were reliably the first to be cut when budgets got tight.
The calculation looks different when modernization measurably lowers resource demand. Then the legacy baggage above stops being an architectural blemish. Then it’s a line item on your procurement list.
Whoever moves to the cloud in 2027 pays for the footprint they bring with them. Whoever cleans up beforehand goes in with a smaller one — and, if they end up staying, has less hardware to procure that may not be available. The measure pays off in both directions. That’s rare enough to make it a priority.
Five questions
The second building block is decoupling the data from the platform it currently sits on. As long as storage characteristics are tied to a specific array, a specific hypervisor, or a specific cloud service, every relocation is a project. And projects take longer than market windows. If you realize in the third quarter of 2027 that you have to move, you won’t be done in the fourth.
You put this layer in before you need it. Whether via a commercial solution, an open-source project, or your own platform’s built-in tooling is the downstream question. The upstream one is how you tell whether it’s any good:
- Are snapshots consistent across sites — across the application group, not per volume?
- Can a backup be restored onto infrastructure other than the one it was created on?
- Does the application move with the data without being modified?
- Does the workload describe its demand, or does the hardware describe its supply?
- When was the last restore onto a foreign platform — not tested, but actually carried out?
The last question is the most uncomfortable, and it’s the only one that counts. All the others can be answered on a slide.
From a position of control
The market, incidentally, won’t help you here. Leading mid-sized IT consultancies grew by an average of 1.9 percent in 2025, after more than ten percent in the years up to 2023. The exceptions are wherever what already exists becomes more efficient: data and AI, cloud transformation, security. No coincidence. In a market without growth, the investment doesn’t go into volume — it goes into freedom of action.
Herbert doesn’t let his Fremen win because they’re stronger. They win because they were the only ones who learned to make do with what the planet gives — and because that’s why they’re the only ones who still have options when it matters.
These are two different deserts. The allocation decides how much Spice you get; over that you have no influence. Your wasteland decides how far it stretches; over that you have all of it.
You can wait until the allocation 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.