Why memory costs what it does

Memory - the RAM in a desktop, and the specialised memory stacked onto AI accelerator cards - is made by a very small number of companies on a very small number of production lines. Those lines are expensive and slow to build, so the world's total output moves slowly even when demand doesn't.

The AI buildout changed demand very quickly. Datacentre hardware consumes memory at a scale desktop computing never has, and it is more profitable per wafer than the memory that goes into consumer machines. Manufacturers have shifted capacity towards it, which leaves less production for ordinary desktop RAM at exactly the moment more people want it.

Less supply, more demand, same factories. Prices go up. Every PC builder in the country is buying from that same squeezed market - the big retailers included, which is why their prices have moved too.

RAM is not a rounding error in a build. It's a real slice of the parts cost, and when that slice goes up the total goes up with it - there is nowhere for it to hide.

It also lands unevenly. On a £3,699 workstation the memory is a modest fraction of the bill. On a £569 office machine it's a much bigger share of a much smaller number, so the cheaper builds feel this hardest. That's the opposite of convenient, and it's worth saying out loud rather than hoping nobody notices.

1. The quote is fixed once you agree it

   Component prices move week to week. Yours doesn't. Once I've confirmed a specification and a price in writing, that's the price - if memory goes up between agreeing and building, that's my problem, not yours.

2. No silent downgrades

   The cheapest way to absorb a component price rise is to quietly fit slower memory, less of it, or a worse power supply, and say nothing. That is how this industry got its reputation. If a part has to change, I ask first and the listing tells you exactly what's in the machine either way.

3. Prices get reviewed, not fudged

   Listed prices are checked against what the parts actually cost. That's why they move - and why I'd rather explain it on a page like this than pretend the number was always going to be that.

Sometimes, yes. If the machine is a want rather than a need and nothing is stopping you waiting, memory has been easing - slowly - and there is a reasonable chance a build costs less later than it does today.

But “slowly” is doing real work in that sentence. This is a supply problem measured in factory build-outs, not weeks, and I'm not going to name you a month. Meanwhile the other parts move on their own schedule: graphics cards, storage and motherboards don't all fall at once, and a saving on memory can be swallowed by a rise somewhere else.

So the straight answer is: if you need a working computer now, buy now and don't agonise about it. If you genuinely don't, tell me what you're after and I'll tell you honestly whether I think waiting is worth it for your budget - including when the answer is “no, this one is fine, buy it”.