The International Data Corporation (IDC) has published a new update to its device market outlook, and the message is blunt: things are getting worse. Under newly-reported pessimistic scenarios, shipments of PCs could shrink by up to 9% in 2026, with a more moderate scenario showing a 5% shrinkage in the market. These figures have been revised from a 2.5% drop, which was recently published in IDC’s November forecast.
Since then, the global memory shortage, which began accelerating in mid-October, has intensified beyond what IDC originally modeled. While the firm isn’t formally rewriting its official forecast entirely, it’s now laying out scenarios that are notably more pessimistic than what it projected just a few weeks ago.
The underlying driver is the same force distorting much of the tech industry in late 2025: AI infrastructure. Memory demand from hyperscalers has surged so aggressively that DRAM and NAND production has been structurally redirected away from consumer devices and toward high-margin enterprise components like high-bandwidth memory and dense DDR5. This is an economically rational choice on the part of memory manufacturers, but IDC is clear that this isn’t a typical boom-and-bust cycle; it’s a strategic reallocation of silicon capacity that could persist for years, not quarters.
I’ll be shocked if it’s not at least double that. Thing is, this is going to be like the covid-19-crypto-bro-GPU-pocalypse that drove up GPU prices so much we now just collectively accept paying double.
Except this time it’s not just GPUs. It’s now hitting RAM, SSDs, HDDs, CPUs, Laptops, probably next year’s smartphones, and who knows what’s next. Motherboards? Power supplies? Cases? Everything else?
It’s not just that these companies are all cannibalizing their consumer capacity for AI customers and it’s not just the hardware they’re buying. As consumer demand plummets because they can no longer afford PCs, companies will reduce consumer production even further because waning demand. It’s a feedback loop whose only killswitch is economic collapse.
Sure we might be able to seek refuge for a while in the secondhand market, but that won’t be our saviour either. As demand increases in the secondhand market, a market with a largely fixed (and likely dwindling) supply, expect sellers to increase their prices too. Whether they’re trying to recoup costs from the first-wave price increase they paid buying new hardware, or just because they know the market can bare inflated prices that are somewhat less inflate compared to new.
And what do we have to look forward to? Prices will “settle” to 2-3x what they were last year compared to 5-6x as today. That’s if there’s a manufacturer left who hasn’t abandoned the consumer segment by that point.
I’m just coming to terms with the fact that the computer I built 2 years ago is probably my last, and my ability to help my neighbours fix their computers probably has a near expiry date. This has been a hobby of mine for decades that the rich have always fought against. With the AI bubble, they may have finally found a way to kill it completely.
I hope I’m wrong but I’m genuinely worried about this. For now I’ll have to wait and see how things go, look to the secondhand market for my next build, and maybe, start learning to solder.
What are you going to solder?
I love soldering but I’ve got no soldering projects at the moment and I’d love some inspiration to attach tiny copper stuff to other tiny copper things.
Also I think that this thing lasting forever depends on some kind of return on ai, which NOBODY is seeing. Current models aren’t accurate or good enough to replace humans in most things yet, and sooner or later the demand for enterprise-capable gpus for this tech is going to dry up. Or there’ll be a new kind of model that will upend the market. Or whatever.
I really don’t think it’s as dire as you make it out to be, but I do think things will be tight for a couple years but I could be wrong.
I think there will be a “return on ai” eventually, just not with this race to the bottom with chatbots. It’ll come from machine learning applied to solving problems in medicine and sciences and other areas of complexity.
Admittedly machine learning and “ai” get lumped together these days but I think there is a difference.
Also, thanks for reading my (long) rant. 😆