Guide · Opinion

RAM & SSDs Are STILL Stocks

The AI memory bubble is leaking. OpenAI is sending all the wrong signals. Prices have somewhat dropped.

7 min read Updated 2026-04-10

Uncertain Demand

I wrote about SSD & RAM prices here: RAM & SSDs are Stocks Now. In that, one of the main points was the "900,000 DRAM wafers per month" — some preliminary agreements with Samsung and SK Hynix that represented roughly 40% of global DRAM production.

So it turns out those were "letters of intent", not contracts. Non-binding. Just vibes, you know how it is.

OpenAI has spent the last month pulling back from its ambitious infrastructure plans. They shut down Sora, the AI video tool that Disney agreed to pour $1 billion into. They walked away from a multi-billion-dollar data centre expansion with Oracle in Texas. The Abilene Stargate campus was supposed to scale from 1.2 gigawatts to 2 gigawatts, and now that expansion is cancelled.

The reasoning behind it shows how reactive they are. OpenAI wants newer chips. The Texas site was being built for Nvidia's Blackwell processors, but by the time power comes online in a year, OpenAI hopes to have access to Nvidia's next-generation hardware elsewhere. So they're abandoning billions in construction because the GPUs will be last season's model.

Also, OpenAI's projected compute spend has dropped from $1.4 trillion by 2033 to around $600 billion by 2030. Still massive. But maybe not "Let me just reserve nearly half the world's DRAM output" massive.

It's all just not very confidence-inspiring.

Why This Doesn't Help You

If you saw headlines about DDR5 prices finally dropping, you might think the crisis is ending. I really don't think it will change anything for the individual consumer.

Prices have fallen (a bit) from their peak. A 32GB DDR5 kit that hit £380 at its peak is now around £280. Somewhere around 25% down. Memory manufacturers' stocks have dropped by similar amounts. OpenAI's scaling back, the market is reacting. Seems like good news for regular people... right?

I don't buy it: £280 is still three, nearly four, times what that kit cost in September 2025 (£80). The drop here is just going from "insane" to "very expensive". I don't see a clear path back to some kind of normal level.

Two things are working against us here.

Efficiency Gets Absorbed, Not Redistributed

In late March, Google published TurboQuant, a compression algorithm that cuts the memory an LLM needs to hold a conversation in progress — the KV cache — by at least 6x. On paper, this is huge. If AI suddenly needs a fraction of the memory it did before, that frees up capacity for everyone else, right?

Imagine being in the boardroom with all the suits when hearing about this. There's not a chance in hell that any of them said "Nice, that will help get consumer prices back in order!" At this point, I'm fairly sure that if someone had said anything like that in one of these rooms, they'd have been thrown out of the window.

When inference suddenly costs less memory, OpenAI doesn't release capacity back to the consumer market. They scale harder (or just take the profit and do nothing). They run more concurrent sessions. They deploy larger models they were holding back. The efficiency gains get absorbed into growth, not redistributed to the rest of us.

This is the pattern with every efficiency improvement in tech. The gains go into expansion, not savings passed downstream.

TurboQuant might reduce pressure on memory supply eventually, in some abstract long-term sense. But I do not believe it will put a cheaper RAM kit in your hands.

The Red Button

Here's the other problem: if demand actually drops enough that prices might fall, memory manufacturers have a lever they can pull: they cut production.

This almost seems too stupid to say, but memory manufacturers like it when memory is expensive; they sell memory...

Anyways, memory is one of the most notoriously cyclical industries in tech, and part of the reason is that Samsung, SK Hynix, and Micron have learned to manage supply. When demand weakens, they don't compete on price, they just throttle output. They've done it before, why not do it again? Literally nobody is going to stop them.

If OPEC+ taught us anything, it's that controlling supply is more profitable than competing. Memory manufacturers, and the semiconductor industry in general, really seem to be getting closer to forming their very own cartel using OPEC+ as their inspiration.

So the scenario where OpenAI's pullback leads to cheap RAM for consumers requires the manufacturers to not respond by tightening supply. That's not how they've historically behaved. I personally cannot see a single reason why they would do that.

Where Prices Actually Are

Looking at the numbers here in the UK: a standard 32GB DDR5 kit peaked around £380 at places like Overclockers UK back in January. Today, it's sitting somewhere between £280 and £300. If you look at the wider European data, prices have stopped climbing sharply, but they are not falling either. They seem to be sitting right where they are.

I think part of what drove the recent drop was speculative unwind. People outside the industry had been hoarding inventory, betting on continued price increases. When prices showed weakness, they panic-sold. Or, in some cases recently, found themselves at a loss after buying a bunch of RAM and hoping to sell it at some ridiculous price later on.

The underlying dynamic, as I see it, has not changed. The capacity that was reallocated to AI hasn't come back. Micron still doesn't sell you RAM anymore.

What This Means for Us

The honest answer is the same as last time: nobody knows where prices are going.

What has changed is that the demand side now looks more uncertain, and it's what prompted me to make this follow-up to my last written piece. OpenAI's purchases were never locked in. Efficiency improvements might reduce how much memory AI actually needs.

What hasn't changed is the supply side. Three companies control the market. All three pivoted away from consumers. All three have the ability to cut production if prices threaten to fall too far. The cartel dynamic doesn't break just because one big customer's cheques bounced.

Prices have (kinda) plateaued. That's not the same as prices recovering. The plateau might be the new floor. Let's hope not.

The advice from the last piece: I have no idea. If anything, there's more uncertainty now, not less. Some might look at this and decide to buy right now, some might hold for next year. See what I mean? It sounds like stock trading...