Here is a claim we are fairly confident about: the reason your next graphics
card costs a small fortune and a memory company’s newest press release have
everything to do with each other, even though one is about gaming and the
other never mentions it once.
On 15 September, Micron showed off the world’s first 512GB DDR5 memory module.
A single stick. Half a terabyte. It runs at 9,200 MT/s, it lets a beefy
dual-socket server pack a frankly silly 12TB of memory, and in the detail that
should make every PC gamer squint, it draws just 16 watts where four of the
older 128GB sticks would pull 44, a more than 60 percent cut in power. It is,
by every engineering measure, a genuinely impressive bit of kit.
It is also not for you. That is the whole point.
Follow the wafers
This is a server part, built for the data centres feeding the AI boom, and it
exists because those data centres have an essentially bottomless appetite for
dense, fast, power-efficient memory. Micron is pouring its cleverest packaging
tricks, in this case stacking DRAM dies vertically and wiring them together
through the silicon, into making memory for those buyers first, because those
buyers pay the most and complain the least.
Gaming does not pay the most. You can guess the rest.
And that is the argument, laid bare. We have
spent weeks now
explaining why GPU prices have gone haywire, and the answer every time comes
back to memory: the same three companies, Samsung, SK Hynix and Micron, make
the chips for both the AI accelerators and your graphics card, and every scrap
of factory capacity they aim at the lucrative server stuff is capacity they
are not aiming at the GDDR7 soldered onto a 5090. A part like this 512GB
monster is not the cause of that squeeze, but it is a perfect, gleaming symbol
of where the whole industry’s attention, and its best engineering, is pointed
right now. And it is not pointed at us.
Why this actually matters for gamers
You could reasonably shrug. Server memory has always been its own world,
priced for corporations, irrelevant to a person building a gaming rig, and for
most of computing history that wall held up just fine, with enterprise gear
off in its own expensive corner and consumer parts trickling down predictably
year after year while nobody had to think about any of it. True, all of that.
But the supply chain does not respect that neat little wall, and it never
really has, because when the single most profitable thing a fab can physically
produce is ultra-dense AI memory, everything downstream feels the pull: the
plain DDR5 in your desktop, the GDDR7 on your GPU, even the price of a humble
stick of RAM you would not have thought twice about two years ago. This
announcement is that whole arms race in miniature, and we are standing several
aisles over in the shop, watching our own components get quietly more
expensive as a side effect of a party we were never invited to.
The one flicker of good news is boring and slow, which is usually how real
good news arrives. Micron says these modules hit full volume production in the
second half of 2027, with AMD and Intel already validating them. More total
memory capacity coming online, eventually, is better than less. It does not
help your wallet this Christmas, or next, but it is at least a sign the
factories are building toward more supply rather than simply cashing in on
scarcity forever.
So no, you cannot buy this, and you would not want to. But if you have been
wondering why the cost of playing games on a PC keeps quietly creeping up
while nobody at the top seems especially bothered, here is your answer,
sitting on a server motherboard, 512 gigabytes at a time. Our
GPU shortlist
still stands: buy for today’s prices, not tomorrow’s promises. (Details via
Tom’s Hardware.)
